geneticprogramming - poznań university of technologygeneticprogramming krzysztofkrawiec...
TRANSCRIPT
Genetic Programming
Krzysztof Krawiec
Laboratory of Intelligent Decision Support SystemsInstitute of Computing Science Poznan University of Technology Poznań Poland
httpwwwcsputpoznanplkkrawiec
June 10 2016
ndash 1
Introduction
Introductionndash 2
Outline and objectives
1 Introduction GP as a variant of EC2 Specific features of GP3 Variants of GP4 Applications5 Some theory6 Case studies
Introductionndash 3
Recommended reading
Koza J R Genetic Programming On the Programming of Computers by Meansof Natural Selection MIT Press 1992
A Field Guide to Genetic Programming (ISBN 978-1-4092-0073-4)httpwwwgp-field-guideorguk
Langdon W B Genetic Programming and Data Structures GeneticProgramming + Data Structures = Automatic Programming Kluwer 1998
Langdon W B amp Poli R Foundations of Genetic Programming Springer-Verlag2002
Riolo R L Soule T amp Worzel B (ed) Genetic Programming Theory andPractice V Springer 2007
Riolo R McConaghy T amp Vladislavleva E (ed) Genetic Programming Theoryand Practice VIII Springer 2010
See httpwwwcsbhamacuk~wblbiblio
Introductionndash 4
Recommended reading
A Field Guide to Genetic Programming httpwwwgp-field-guideorguk[39]
(This presentation uses some figures from the Field Guide)
Introductionndash 5
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Introduction
Introductionndash 2
Outline and objectives
1 Introduction GP as a variant of EC2 Specific features of GP3 Variants of GP4 Applications5 Some theory6 Case studies
Introductionndash 3
Recommended reading
Koza J R Genetic Programming On the Programming of Computers by Meansof Natural Selection MIT Press 1992
A Field Guide to Genetic Programming (ISBN 978-1-4092-0073-4)httpwwwgp-field-guideorguk
Langdon W B Genetic Programming and Data Structures GeneticProgramming + Data Structures = Automatic Programming Kluwer 1998
Langdon W B amp Poli R Foundations of Genetic Programming Springer-Verlag2002
Riolo R L Soule T amp Worzel B (ed) Genetic Programming Theory andPractice V Springer 2007
Riolo R McConaghy T amp Vladislavleva E (ed) Genetic Programming Theoryand Practice VIII Springer 2010
See httpwwwcsbhamacuk~wblbiblio
Introductionndash 4
Recommended reading
A Field Guide to Genetic Programming httpwwwgp-field-guideorguk[39]
(This presentation uses some figures from the Field Guide)
Introductionndash 5
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Outline and objectives
1 Introduction GP as a variant of EC2 Specific features of GP3 Variants of GP4 Applications5 Some theory6 Case studies
Introductionndash 3
Recommended reading
Koza J R Genetic Programming On the Programming of Computers by Meansof Natural Selection MIT Press 1992
A Field Guide to Genetic Programming (ISBN 978-1-4092-0073-4)httpwwwgp-field-guideorguk
Langdon W B Genetic Programming and Data Structures GeneticProgramming + Data Structures = Automatic Programming Kluwer 1998
Langdon W B amp Poli R Foundations of Genetic Programming Springer-Verlag2002
Riolo R L Soule T amp Worzel B (ed) Genetic Programming Theory andPractice V Springer 2007
Riolo R McConaghy T amp Vladislavleva E (ed) Genetic Programming Theoryand Practice VIII Springer 2010
See httpwwwcsbhamacuk~wblbiblio
Introductionndash 4
Recommended reading
A Field Guide to Genetic Programming httpwwwgp-field-guideorguk[39]
(This presentation uses some figures from the Field Guide)
Introductionndash 5
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Recommended reading
Koza J R Genetic Programming On the Programming of Computers by Meansof Natural Selection MIT Press 1992
A Field Guide to Genetic Programming (ISBN 978-1-4092-0073-4)httpwwwgp-field-guideorguk
Langdon W B Genetic Programming and Data Structures GeneticProgramming + Data Structures = Automatic Programming Kluwer 1998
Langdon W B amp Poli R Foundations of Genetic Programming Springer-Verlag2002
Riolo R L Soule T amp Worzel B (ed) Genetic Programming Theory andPractice V Springer 2007
Riolo R McConaghy T amp Vladislavleva E (ed) Genetic Programming Theoryand Practice VIII Springer 2010
See httpwwwcsbhamacuk~wblbiblio
Introductionndash 4
Recommended reading
A Field Guide to Genetic Programming httpwwwgp-field-guideorguk[39]
(This presentation uses some figures from the Field Guide)
Introductionndash 5
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Recommended reading
A Field Guide to Genetic Programming httpwwwgp-field-guideorguk[39]
(This presentation uses some figures from the Field Guide)
Introductionndash 5
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Background
Backgroundndash 6
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Evolutionary Computation (EC)
Heuristic bio-inspired global search algorithms
Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the fitness function fbeing optimized
Formulation
plowast = argmaxpisinP
f (p)
where
P is the considered space (search space) of candidate solutions (solutions forshort)
f is a (maximized) fitness function
plowast is an optimal solution (an ideal) that maximizes f
Backgroundndash 7
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Generic evolutionary algorithm
Evolutionary Algorithm
Population P of individuals
Evaluation
Selection
Mutation and recombination
Initialization of population P
Solutionindividual s
f(s)
Output Best solution s+
Termination criteria
Fitness function f
Backgroundndash 8
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
[Unique] characteristic of EC
Black-box optimization (f primes dependency on the independent variables does nothave to be known or meet any criteria)
Variables do not have to be explicitly defined
Fining an optimum cannot be guaranteed but in practice a well-performingsuboptimal solution is often satisfactory
Importance of crossover a recombination operator that makes the solutionsexchange certain elements (variable values features)
Without crossover EC boils down parallel stochastic local search
Backgroundndash 9
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
What is genetic programming
What is genetic programmingndash 10
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Genetic programming
In a nutshell
A variant of EC where the genotypes represent programs ie entities capable ofreading in input data and producing some output data in response to that input
Fitness function f measures the similarity of the output produced by the programto the desired output given as a part of task statement
Standard representation expression trees
Important implication Additional input required by the algorithm (compared to EC)
Set of instructions (programming language of consideration)
Data to run the programs on
What is genetic programmingndash 11
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Conceptual preliminaries of GP Space of candidate solutions P
Candidate solutions p isin P evolving under the selection pressure of the fitnessfunction f are themselves functions of the form p I rarrO
I and O are respectively the spaces of input data and output data accepted andproduced by programs from P
Cardinality of |P| is typically large or infinite
The set of program inputs I even if finite is usually so large that running eachcandidate solution on all possible inputs becomes intractable
GP algorithms typically evaluate solutions on a sample I prime sub I |I prime| |I | of possibleinputs and fitness is only an approximate estimate of solution quality
The task is given as a set of fitness cases ie pairs (xi yi ) isin I timesO where xiusually comprises one or more independent variables and yi is the output variable
What is genetic programmingndash 12
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Conceptual preliminaries Fitness function f
In most cases (and most real-world applications of GP) fitness function fmeasures the similarity of the output produced by the program to the desiredoutput given as a part of task statement
Then fitness can be expressed as a monotonous function of the divergence ofprogramrsquos output from the desired one for instance as
f (p) =minussumi
||yi minusp(xi )|| (1)
where
p(xi ) is the output produced by program p for the input data xi
|| middot || is a metric (a norm) in the output space O
i iterates over all fitness cases
What is genetic programmingndash 13
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Conceptual preliminaries Character of candidate solutions
The candidate solutions in GP are being assembled from elementary entities calledinstructions
A part of formulation of a GP task is then also an instruction set I ie a set ofsymbols used by the search algorithm to compose the programs (candidatesolutions)
Design of I usually requires some background knowledgeIn particular it should comprise all instructions necessary to find solution to theproblem posed (closure)
What is genetic programmingndash 14
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Genetic programming
Main evolution loop (lsquovanilla GPrsquo)
1 procedure GeneticProgramming(f I ) f - fitness function I - instruction set2 P larrplarrRandomProgram(I ) Initialize population3 repeat Main loop over generations4 for p isinP do Evaluation5 pf larr f (p) pf is a lsquofieldrsquo in program p that stores its fitness6 end for7 P prime larr 0 Next population8 repeat Breeding loop9 p1 larrTournamentSelection(P) First parent10 p2 larrTournamentSelection(P) Second parent11 (o1o2)larrCrossover(p1 p2)12 o1 larrMutation(o1 I )13 o2 larrMutation(o2 I )14 P prime larrP prime cupo1o215 until |P prime|= |P|16 P larrP prime
17 until StoppingCondition(P)18 return argmaxpisinP pf
19 end procedure
What is genetic programmingndash 15
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Crossover
Crossover exchange of randomly selected subexpressions (subtree swapping crossover)
1 function Crossover(p1p2)2 repeat3 s1 larr Random node in p14 s2 larr Random node in p25 (pprime1p
prime2)larr Swap subtrees rooted in s1 and s2
6 until Depth(pprime1)lt dmax andDepth(pprime2)lt dmax dmax is the tree depthlimit
7 return (pprime1pprime2)
8 end function
What is genetic programmingndash 16
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Mutation
Mutation replace a randomly selected subexpression with a new randomly generatedsubexpression
1 function Mutation(pI )2 repeat3 slarr Random node in p4 s prime larrRandomProgram(I )5 pprime larr Replace the subtree rooted in s with s prime
6 until Depth(pprime)lt dmax dmax is the tree depth limit7 return pprime
8 end function
What is genetic programmingndash 17
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Exemplary run Setup
Objective Find program whose output matches x2+x+1 over the range [minus11]Such tasks can be considered as a form of regressionAs solutions are built by manipulating code (instructions) this is referred to assymbolic regression
Fitness sum of absolute errors for x isin minus10minus09 0910In other words the set of fitness cases is
xi -10 -09 0 09 10yi 1 091 1 271 3
What is genetic programmingndash 18
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Exemplary run Setup
Instruction setNonterminal (function) set + - (protected division) and x all operating onfloatsTerminal set x and constants chosen randomly between -5 and +5
Selection fitness proportionate (roulette wheel) non elitist
Initial pop ramped half-and-half (depth 1 to 2 50 of terminals are constants)(to be explained later)
Parameterspopulation size 450 subtree crossover25 reproduction25 subtree mutation no tree size limits
Termination when an individual with fitness better than 01 found
What is genetic programmingndash 19
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Initial population (population 0)
What is genetic programmingndash 20
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Fitness assignment for population 0
Fitness values f(a)=77 f(b)=110 f(c)=1798 f(d)=287
What is genetic programmingndash 21
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Breeding
Assume
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over
What is genetic programmingndash 22
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Population 1
Population 0
Population 1
Individual d in population 1 has fitness 0What is genetic programmingndash 23
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Summary of our first glimpse at GP
Summary of our first glimpse at GPndash 24
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Specific features of GP
The solutions evolving under the selection pressure of the fitness function arethemselves functions (programs)
GP operates on symbolic structures of varying lengthsThere are no variables for the algorithm to operate on (at least in the commonsense)
The program can be tested only on a limited number of fitness cases (tests)
=rArr In contrast to most EC methods that are typically placed in optimizationframework GP is by nature an inductive learning approach that fits into the domain ofmachine learning []
Summary of our first glimpse at GPndash 25
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
In a broader context
As opposed to typical ML approaches GP is very genericArbitrary programming language arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve asmeans to provide the algorithm with prior knowledge
(common sense knowledge background knowledge domain knowledge)
GP is not the only approach to program induction (but probably the best one )See eg inductive logic programming ILP
GP embodies the ultimate goal of AI to build a system capable ofself-programming (adaptation learning)
Summary of our first glimpse at GPndash 26
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
What is GP ndash Question revisited
Genetic programming is a branch of computer science studying heuristicalgorithms based on neo-Darwinian principles for synthesizing programs iediscrete symbolic compositional structures that process data
Summary of our first glimpse at GPndash 27
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Why should GP be considered a viable approach to program synthesis
GP combines two powerful concepts marked in underline in the above definition1 Representing candidate solutions as programs
which in general can conduct any Turing-complete computation (egclassification regression clustering reasoning problem solving etc) and thusenable capturing solutions to any type of problems (whether the task is eglearning optimization problem solving game playing etc)
2 Searching the space of candidate solutions using the lsquomechanicsrsquo borrowed frombiological evolutionwhich is unquestionably a very powerful computing paradigm given that itresulted in life on Earth and development of intelligent beings
Summary of our first glimpse at GPndash 28
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Consequences of the above definition
Heuristic nature of search
Symbolic program representation
Input sensitivity and inductive character
State-fullness
Unconstrained data types
Summary of our first glimpse at GPndash 29
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Origins of GP
httpwwwgenetic-programmingcomjohnkozahtml
Home assignment Indentify John Kozarsquos unique visual trait )
Summary of our first glimpse at GPndash 30
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Life demonstration of GP using ECJ
Life demonstration of GP using ECJndash 31
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
GP on ECJ
ECJ Evolutionary Computation in Javahttpcsgmuedu~eclabprojectsecj
by Sean Luke Liviu Panait Gabriel Balan Sean Paus Zbigniew Skolicki ElenaPopovici Keith Sullivan et al
Probably the most popular freely available framework for EC with a strongsupport for GP
Licensed under Academic Free License version 30
As of March 2012 version 20
Many other libraries integrate with ECJ
Life demonstration of GP using ECJndash 32
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Selected ECJ features
GUI with charting
Platform-independent checkpointing and logging
Hierarchical parameter files
Multithreading
Mersenne Twister Random Number Generators (compare tohttpwwwalifecouknonrandom)
Abstractions for implementing a variety of EC forms
Prepared to work in a distributed environment (including so-called island model)
Life demonstration of GP using ECJndash 33
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
GP-related ECJ features
GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Defined Functions and Automatically Defined Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant regression multiplexerlawnmower parity two-box edge serengeti)
Life demonstration of GP using ECJndash 34
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Exemplary run of ECJ
Standard outputjava ecEvolve -file ecappregressionquinticercparamsThreads breed1 eval1Seed 1427743400Job 0Setting upProcessing GP TypesProcessing GP Node ConstraintsProcessing GP Function SetsProcessing GP Tree Constraints-01306332228659439200164875774146594280653340439694114301402200189629743-0037506348565697010001402771209365470606602806044824949013869498395598084Initializing Generation 0Subpop 0 best fitness of generation Fitness Standardized=11303205 Adjusted=046941292 Hits=10Generation 1Subpop 0 best fitness of generation Fitness Standardized=06804932 Adjusted=059506345 Hits=7
Life demonstration of GP using ECJndash 35
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Exemplary run The result
The log file produced by the runGeneration 0Best IndividualSubpopulation 0Evaluated trueFitness Standardized=11303205 Adjusted=046941292 Hits=10Tree 0( (sin ( x x)) (cos (+ x x)))Generation 1Best IndividualSubpopulation 0Evaluated trueFitness Standardized=06804932 Adjusted=059506345 Hits=7Tree 0( (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos ( x x))) (- x x))))
Life demonstration of GP using ECJndash 36
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Exemplary run
The log file produced by the runBest Individual of RunSubpopulation 0Evaluated trueFitness Standardized=008413165 Adjusted=092239726 Hits=17Tree 0( ( ( (- ( ( ( ( x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp ( x( ( (- ( ( ( ( x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp ( x (sin x))))(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos ( ( (- ( ((exp (rlog x)) (+ x ( ( (exp (rlog x))(rlog x)) x))) (exp ( ( ( (- (exp (rlogx)) x) (rlog x)) x) (sin ( x x))))) (sinx)) ( x ( ( (- ( ( ( ( x x) (rlogx)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-x x))) (exp ( x (sin x)))) (sin x)) (rlogx)) (exp (rlog x))))) x))))
Life demonstration of GP using ECJndash 37
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
GP in R
httpcranr-projectorgwebpackagesgprindexhtml
Life demonstration of GP using ECJndash 38
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
GP in Mathematica
Life demonstration of GP using ECJndash 39
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Assessment of GP techniques
Assessment of GP techniquesndash 40
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Criteria for assessing the quality of GP-evolved solutions
Criteria for assessing GP algorithmssuccess rate (percentage of evolutionary runs ended with success)
time-to-success (can be infin)
error of the best-of-run individual
Criteria for assessing programs obtained with GP
error rate (percentage of tests passed)
program size (number of instructions)
execution time
transparency
Assessment of GP techniquesndash 41
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
GP Benchmarks
Problem Definition (formula)Sextic x6minus2x4+x2
Septic x7minus2x6+x5minusx4+x3minus2x2+xNonic x9+x8+x7+x6+x5+x4+x3+x2+xR1 (x+1)3(x2minusx+1)R2 (x5minus3x3+1)(x2+1)R3 (x6+x5)(x4+x3+x2+x+1)
Assessment of GP techniquesndash 42
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Even more GP benchmarks
Symbolic Regression Drug Bioavailability [43]
Tower [52] Protein Structure Classification [58]
Boolean Functions Time Series Forecasting [53]
N-Multiplexer [17] N-Majority [17] N-Parity [17] Path-finding and Planning
Generalised Boolean Circuits [12 59] Physical Travelling Salesman [28]
Digital Adder [55] Artificial Ant [17]
Order [9] Lawnmower [18]
Digital Multiplier [55] Tartarus Problem [6]
Majority [9] Maximum Overhang [36]
Classification Circuit Design [29]
mRNA Motif Classification [24] Control Systems
DNA Motif Discovery [25] Chaotic Dynamic Systems Control [26]
Intrusion Detection [11] Pole Balancing [31]
Protein Classification [19] Truck Control [16]
Intertwined Spirals [17]
Predictive Modelling
Mackey-Glass Chaotic Time Series [21]
Financial Trading [5 4 8]
Sunspot Prediction [17]
GeneChip Probe Performance [22]
Prime Number Prediction [54]
Assessment of GP techniquesndash 43
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
and more
Game-Playing
TORCS Car Racing [50]
Ms PacMan [10]
Othello [27]
Chessboard Evaluation [44]
Backgammon [44]
Mario [49]
NP-Complete Puzzles [14]
Robocode [44]
Rush Hour [44]
Checkers [44]
Freecell [44]
Dynamic Optimisation
Dynamic Symbolic Regression [34 35 51]
Dynamic Scheduling [13]
Traditional Programming
Sorting [15 1]
Assessment of GP techniquesndash 44
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Case study
Case studyndash 45
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Case study 1 Evolution of temperature models
Based on Karolina Stanisławska Krzysztof Krawiec Zbigniew W KundzewiczModeling Global Temperature Changes using Genetic Programming ndash A Case Study
Institute of Computing Science Poznan University of Technology Poznan Poland
Institute for Agricultural and Forest Environment Polish Academy of SciencesPoznan Poland and Potsdam Institute for Climate Impact Research PotsdamGermany
Konferencja Algorytmoacutew Ewolucyjnych i Optymalizacji Globalnej KAEiOGSeptember 21 2011
Case studyndash 46
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
A more detailed view on GP(vanilla GP is not the whole story)
A more detailed view on GP (vanilla GP is not the whole story)ndash 47
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Where to get [random] candidate solutions from
Every stochastic search method relies on some sampling algorithm(s)
The distribution of randomly generated solutions is important as it impliescertain bias of the algorithm
ProblemsWe donrsquot know the lsquoidealrsquo distribution of GP programsEven if we knew it it may be difficult to design an algorithm that obeys it
The most widely used contemporary initialization methods take care only of thesyntax of generated programs
Mainly height constraint
A more detailed view on GP (vanilla GP is not the whole story)ndash 48
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Initialization Full method
Specify the maximum tree height hmax
The full method for initializing treesChoose nonterminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 49
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Initialization Grow method
Specify the maximum tree height hmax
The grow method for initializing treesChoose nonterminal or terminal nodes at random until hmax is reachedThen choose only from terminals
A more detailed view on GP (vanilla GP is not the whole story)ndash 50
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Homologous crossover for GP
Earliest example one-point crossover [23] identify a common region in theparents and swap the corresponding trees
The common region is the lsquointersectionrsquo of parent trees
A more detailed view on GP (vanilla GP is not the whole story)ndash 51
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Uniform crossover for GP
Works similarly to uniform crossover in GAs
The offspring is build by iterating over nodes in the common region and flipping acoin to decide from which parent should an instruction be copied [38]
A more detailed view on GP (vanilla GP is not the whole story)ndash 52
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
How to employ multiple operators for lsquobreedingrsquo
How should the particular operators coexist in an evolutionary process In other words
How should they be superimposed
What should be the lsquopipingrsquo of particular breeding pipelines
A topic surprisingly underexplored in GP (and in EC probably too)
An example Which is betterpopsubpop0speciespipe = ecgpkozaMutationPipelinepopsubpop0speciespipenum-sources = 1popsubpop0speciespipesource0 = ecgpkozaCrossoverPipeline
Orpopsubpop0speciespipenum-sources = 2popsubpop0speciespipesource0 = ecgpkozaCrossoverPipelinepopsubpop0speciespipesource0prob = 09popsubpop0speciespipesource1 = ecgpkozaMutationPipelinepopsubpop0speciespipesource1prob = 01
A more detailed view on GP (vanilla GP is not the whole story)ndash 53
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
The Challenges for GP
The Challenges for GPndash 54
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Bloat
The evolving expressions tend to grow indefinitely in sizeFor tree-based representations this growth is typically exponential[-ish]
Evaluation becomes slow algorithm stalls memory overrun likely
One of the most intensely studied topics in GP 240+ papers as of March 2012
The Challenges for GPndash 55
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Bloat example
Average number of nodes per generation in a typical run of GP solving the Sexticproblem x6minus2x4+x2
(GP dotted line)
The Challenges for GPndash 56
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Countermeasures for bloat
Constraining tree heightSurprisingly can speed up bloat
Favoring small programsLexicographic parsimony pressure given two equally fit individuals prefer (select)the one represented by a smaller tree
Bloat-aware operators size-fair crossover
The Challenges for GPndash 57
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Highly non-uniform distribution of program lsquobehaviorsrsquo
Convergence of binary Boolean random linear functions (composed of AND NANDOR NOR 8 bits)
From [20] Langdon W B Cantuacute-Paz E (ed) Random Search is Parsimonious LateBreaking Papers at the Genetic and Evolutionary Computation Conference(GECCO-2002) AAAI 2002 308-315
The Challenges for GPndash 58
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
High cost of evaluation
Running a program on multiple inputscan be expensive
Particularly for some types of dataeg images
Solutions
Caching of outcomes of subprograms
Parallel execution of programs onparticular fitness cases
Bloat prevention methods
The Challenges for GPndash 59
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Variants of GP
Variants of GPndash 60
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Strongly typed GP (STGP)
A way to incorporate prior knowledge and impose a structure on programs [30]
ImplementationProvide a set of typesFor each instruction define the types of its arguments and outcomesMake the operators type-aware
Mutation substitute a random tree of a proper typeCrossover swap trees of compatible1 types
1Compatible belonging to the same lsquoset typersquoVariants of GPndash 61
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Strongly typed GP in ECJ
For the problem of simple classifiers represented as decision trees
Classifier syntaxClassifier = Class_idClassifier = if_then_else(Condition ClassifierClassifier)Condition = Input_Variable = Constant_Value
Implementation in ECJ parameter filesgptypeasize = 3gptypea0name = classgptypea1name = vargptypea2name = constgptypessize = 0gptcsize = 1gptc0 = ecgpGPTreeConstraintsgptc0name = tc0gptc0fset = f0gptc0returns = class
gpncsize = 4gpnc0 = ecgpGPNodeConstraintsgpnc0name = ncSimpleClassifiergpnc0returns = classgpnc0size = 0gpnc1 = ecgpGPNodeConstraintsgpnc1name = ncCompoundClassifiergpnc1returns = classgpnc1size = 4gpnc1child0 = vargpnc1child1 = constgpnc1child2 = classgpnc1child3 = classgpnc2 = ecgpGPNodeConstraintsgpnc2name = ncVariablegpnc2returns = vargpnc2size = 0gpnc3 = ecgpGPNodeConstraintsgpnc3name = ncConstantgpnc3returns = constgpnc3size = 0
Variants of GPndash 62
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Linear Genetic Programming
MotivationTree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
ProsDirectly portable to machine code fast executionNatural correspondence to standard (GA-like) crossover operator
Applications direct evolution of machine code [32]
Initial register contents
Final register contents
x1
x2 O1 O2
x3
O3 O4 g2
g3
g1r1
r2
r3
r1
r2
r3
Variants of GPndash 63
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Linear GP
Variants of GPndash 64
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Stack-based GP
The best-known representative Push and PushGPhampshireedulspectorpushhtml [48]
ProsVery simple syntax program = instruction | literal | ( program )No need to specify the number of registersThe top element of a stack has the natural interpretation of program outcomeNatural possibility of implementing autoconstructive programs [47]Includes certain features that make it Turing-complete (eg YANK instruction)
Variants of GPndash 65
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Push Example
Program
( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )
Initial stack states
BOOLEAN STACK ()CODE STACK ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR )FLOAT STACK ()INTEGER STACK ()
Stack states after program execution
BOOLEAN STACK ( TRUE )CODE STACK ( ( 2 3 INTEGER 41 52 FLOAT+ TRUE FALSE BOOLEANOR ) )FLOAT STACK ( 93 )INTEGER STACK ( 6 )
httphampshireedulspectorpush3-descriptionhtml
Variants of GPndash 66
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Other variants of GP
Grammatical Evolution The grammar of the programming language ofconsideration is given as input to the algorithm Individuals encode the choice ofproductions in the derivation tree (which of available alternative productionshould be chosen modulo the number of productions available at given step ofderivation)
Graph-based GPMotivation standard GP cannot reuse subprograms (within a single program)Example Cartesian Genetic Programming
Variants of GPndash 67
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Other variants of GP
Multiobjective GP The extra objectives canCome with the problemResult from GPrsquos specifics eg use program size as the second (minimized)objectiveBe associated with different tests (eg feature tests [40])
Developmental GP (eg using Push)
Probabilistic GP (a variant of EDA Estimation of Distribution Algorithms)The algorithm maintains a probability distribution P instead of a populationIndividuals are generated from P lsquoon demandrsquoThe results of individualsrsquo evaluation are used to update P
Variants of GPndash 68
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Simple EDA-like GP PIPE
Probabilistic Incremental Program Evolution [41]
Variants of GPndash 69
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Selected theoretical results
Selected theoretical resultsndash 70
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Schemata theorem for GP
Exact formula for the expected number of individuals sampling a schema a thenext generation [37]
Plus later work for other types of crossover
Selected theoretical resultsndash 71
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Semantic crossover
Selected theoretical resultsndash 72
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Applications of GP
Applications of GPndash 73
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Review
GP produced a number of solutions that are human-competitive ie a GPalgorithm automatically solved a problem for which a patent exists2
A recent award-winning work has demonstrated the ability of a GP system toautomatically find and correct bugs in commercially-released software whenprovided with test data3
GP is one of leading methodologies that can be used to lsquoautomatersquo sciencehelping the researchers to find the hidden complex patterns in the observedphenomena4
2Koza J R Keane M A Streeter M J Mydlowec W Yu J Lanza G 2003 Genetic Pro-gramming IV Routine Human-Competitive Machine Intelligence Kluwer Academic Publishers
3Arcuri A Yao X A novel co-evolutionary approach to automatic software bug fixing In WangJ (Ed) 2008 IEEE World Congress on Computational Intelligence IEEE Computational IntelligenceSociety IEEE Press Hong Kong
4Schmidt M Lipson H 3 Apr 2009 Distilling free-form natural laws from experimental dataScience 324 (5923) 81ndash85
Applications of GPndash 74
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Humies
() Entries were solicited for cash awards for human-competitive results that wereproduced by any form of genetic and evolutionary computation and that were published
Applications of GPndash 75
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Humies
The conditions to qualify(A) The result was patented as an invention in the past is an improvement over apatented invention or would qualify today as a patentable new invention(B) The result is equal to or better than a result that was accepted as a new scientificresult at the time when it was published in a peer-reviewed scientific journal(C) The result is equal to or better than a result that was placed into a database orarchive of results maintained by an internationally recognized panel of scientificexperts(D) The result is publishable in its own right as a new scientific result mdash independentof the fact that the result was mechanically created(E) The result is equal to or better than the most recent human-created solution to along-standing problem for which there has been a succession of increasingly betterhuman-created solutions(F) The result is equal to or better than a result that was considered an achievementin its field at the time it was first discovered(G) The result solves a problem of indisputable difficulty in its field(H) The result holds its own or wins a regulated competition involving humancontestants (in the form of either live human players or human-written computerprograms)
Applications of GPndash 76
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Selected Gold Humies
2004 Jason D Lohn Gregory S Hornby Derek S Linden NASA Ames ResearchCenterAn Evolved Antenna for Deployment on NASArsquos Space Technology 5 Mission
httpidesignucscedupapershornby_ec11pdf
Applications of GPndash 77
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Selected Gold Humies using GP
2009 Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley WeimerAutomatically finding patches using genetic programming A GeneticProgramming Approach to Automated Software Repair
Applications of GPndash 78
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Selected Gold Humies using GP
2008 Lee Spector David M Clark Ian Lindsay Bradford Barr Jon KleinGenetic Programming for Finite Algebras
2010 Natalio Krasnogor Paweł Widera Jonathan GaribaldiEvolutionary design of the energy function for protein structure prediction GPchallenge evolving the energy function for protein structure prediction Automateddesign of energy functions for protein structure prediction by means of geneticprogramming and improved structure similarity assessment
2011 Achiya Elyasaf Ami Hauptmann Moshe SipperGA-FreeCell Evolving Solvers for the Game of FreeCell
Applications of GPndash 79
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Other applications
classification problems in machine learning [] object recognition [ 33] orlearning game strategies [] [2 57] has demonstrated the ability of a GP system to automatically find andcorrect bugs in commercially-released software when provided with test data
In the context of this paper it deserves particular attention that GP is one ofleading methodologies that can be used to lsquoautomatersquo science helping theresearchers to find the hidden complex patterns in the observed phenomena
In this spirit in their seminal paper [42] have shown how GP can be used toinduce scientific laws from experimental data Many other studies havedemonstrated the usefulness of GP for modeling different phenomena includingthose of natural origins [45 3 7 56 46]
See [] for an extensive review of GP applications
Applications of GPndash 80
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Case study 2 Evolution of feature detectors
Based on Krzysztof Krawiec Bartosz Kukawka and Tomasz Maciejewski (2010)Evolving cascades of voting feature detectors for vehicle detection in satellite imageryIn IEEE Congress on Evolutionary Computation (CEC 2010) Barcelona IEEE Presspages 2392-2399
Applications of GPndash 81
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Additional resources
Additional resourcesndash 82
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Additional resources
Evolutionary Computation in Java csgmuedu~eclabprojectsecjGeneric software framework for EA well-prepared to work with GP
The online GP bilbiography wwwcsbhamacuk~wblbiblio
The genetic programming lsquohome pagersquo (a little bit messy but still valuable)httpwwwgenetic-programmingcom
Additional resourcesndash 83
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
Bibliography
Bibliographyndash 84
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
A Agapitos and S M LucasEvolving Modular Recursive Sorting AlgorithmsIn Proc EuroGP 2007
A Arcuri and X YaoA novel co-evolutionary approach to automatic software bug fixingIn J Wang editor 2008 IEEE World Congress on Computational Intelligencepages 162ndash168 Hong Kong 1-6 June 2008 IEEE Computational IntelligenceSociety IEEE Press
M Arganis R Val J Prats K Rodriguez R Dominguez and J DolzGenetic programming and standardization in water temperature modellingAdvances in Civil Engineering 2009 2009
A Brabazon M OrsquoNeill and I DempseyAn Introduction to Evolutionary Computation in FinanceIEEE Computational Intelligence Magazine 3(4)42ndash55 2008
R Bradley A Brabazon and M OrsquoNeillDynamic High Frequency Trading A Neuro-Evolutionary ApproachIn Proc EvoWorkshops 2009
G Cuccu and F GomezWhen novelty is not enoughIn Proc EvoApplications 2011
Bibliographyndash 85
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
M Daga and M C DeoAlternative data-driven methods to estimate wind from waves by inversemodelingNatural Hazards 49(2)293ndash310 May 2009
I Dempsey M OrsquoNeill and A BrabazonAdaptive Trading With Grammatical EvolutionIn Proc CEC 2006
G Durrett F Neumann and U-M OrsquoReillyComputational Complexity Analysis of Simple Genetic Programming On TwoProblems Modeling Isolated Program SemanticsIn Proc FOGA 2011
E Galvaacuten-Loacutepez J Swafford M OrsquoNeill and A BrabazonEvolving a Ms PacMan Controller Using Grammatical EvolutionIn Applications of Evolutionary Computation Springer 2010
J V Hansen P B Lowry R D Meservy and D M McDonaldGenetic Programming for Prevention of Cyberterrorism through Dynamic andEvolving Intrusion DetectionDecision Support Systems 431362ndash1374 2007
Bibliographyndash 86
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
S Harding J F Miller and W BanzhafDevelopments in Cartesian Genetic Programming self-modifying CGPGPEM 11397ndash439 2010
D Jakobović and L BudinDynamic Scheduling with Genetic ProgrammingIn Proc EuroGP 2006
G Kendall A Parkes and K SpoererA Survey of NP-Complete PuzzlesInternational Computer Games Association Journal 31(1)13ndash34 2008
K E Kinnear JrEvolving a Sort Lessons in Genetic ProgrammingIn Proc of the International Conference on Neural Networks 1993
J KozaA Genetic Approach to the Truck Backer Upper Problem and the Inter-twinedSpiral ProblemIn Proc International Joint Conference on Neural Networks 1992
J R KozaGenetic Programming On the Programming of Computers by Means of NaturalSelectionMIT Press Cambridge MA USA 1992
Bibliographyndash 87
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
J R KozaGenetic Programming II Automatic Discovery of Reusable ProgramsMIT Press Cambridge Massachusetts May 1994
W Langdon and W BanzhafRepeated Patterns in Genetic ProgrammingNatural Computing 7589ndash613 2008
W B LangdonRandom search is parsimoniousIn E Cantuacute-Paz editor Late Breaking Papers at the Genetic and EvolutionaryComputation Conference (GECCO-2002) pages 308ndash315 New York NY 9-13July 2002 AAAI
W B Langdon and W BanzhafRepeated Sequences in Linear Genetic Programming GenomesComplex Systems 15(4)285ndash306 2005
W B Langdon and A P HarrisonEvolving Regular Expressions for GeneChip Probe Performance PredictionIn Proc PPSN pages 1061ndash1070 2008
W B Langdon and R PoliFoundations of Genetic ProgrammingSpringer-Verlag 2002
Bibliographyndash 88
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
W B Langdon J Rowsell and A P HarrisonCreating Regular Expressions as mRNA Motifs with GP to Predict Human ExonSplittingIn Proc GECCO 2009
W B Langdon O Sanchez Graillet and A P HarrisonAutomated DNA Motif DiscoveryarXivorg 2010
M Lones A Tyrrell S Stepney and L CavesControlling Complex Dynamics with Artificial Biochemical NetworksIn Proc EuroGP pages 159ndash170 2010
S LucasOthello Competitionhttpprotectkern-1667emrelaxalgovalessexacuk8080othellohtmlOthellohtml 2012[Online accessed 27-Jan-2012]
Bibliographyndash 89
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
S LucasThe Physical Travelling Salesperson Problemhttpprotectkern-1667emrelaxalgovalessexacukptspptsphtml2012[Online accessed 27ndashJan-2012]
T McConaghyFFX Fast Scalable Deterministic Symbolic Regression TechnologyIn Proc GPTP 2011
D J MontanaStrongly typed genetic programmingBBN Technical Report 7866 Bolt Beranek and Newman Inc 10 MoultonStreet Cambridge MA 02138 USA 7 May 1993
M Nicolau M Schoenauer and W BanzhafEvolving Genes to Balance a PoleIn Proc EuroGP 2010
Bibliographyndash 90
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
P Nordin and W BanzhafGenetic programming controlling a miniature robotIn E V Siegel and J R Koza editors Working Notes for the AAAI Symposiumon Genetic Programming pages 61ndash67 MIT Cambridge MA USA 10ndash12 Nov1995 AAAI
G Olague and L TrujilloEvolutionary-computer-assisted design of image operators that detect interestpoints using genetic programmingImage and Vision Computing 29(7)484ndash498 2011
M OrsquoNeill A Brabazon and E HembergSubtree Deactivation Control with Grammatical Genetic Programming inDynamic EnvironmentsIn Proc CEC 2008
M OrsquoNeill and C RyanGrammatical Evolution by Grammatical Evolution The Evolution of Grammarand Genetic CodeIn Proc EuroGP pages 138ndash149 Springer-Verlag 5ndash7 Apr 2004
M Paterson Y Peres M Thorup P Winkler and U ZwickMaximum OverhangIn Proc 19th Annual ACM-SIAM Symposium on Discrete Algorithms 2008
Bibliographyndash 91
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
R PoliExact schema theory for genetic programming and variable-length geneticalgorithms with one-point crossoverGenetic Programming and Evolvable Machines 2(2)123ndash163 June 2001
R Poli and W B LangdonOn the search properties of different crossover operators in genetic programmingIn J R Koza W Banzhaf K Chellapilla K Deb M Dorigo D B Fogel M HGarzon D E Goldberg H Iba and R Riolo editors Genetic Programming1998 Proceedings of the Third Annual Conference pages 293ndash301 University ofWisconsin Madison Wisconsin USA 22-25 July 1998 Morgan Kaufmann
R Poli W B Langdon and N F McPheeA field guide to genetic programmingPublished via httplulucom and freely available athttpwwwgp-field-guideorguk 2008(With contributions by J R Koza)
B J Ross and H ZhuProcedural texture evolution using multiobjective optimizationNew Generation Computing 22(3)271ndash293 2004
Bibliographyndash 92
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
R P Salustowicz and J SchmidhuberProbabilistic incremental program evolutionEvolutionary Computation 5(2)123ndash141 1997
M Schmidt and H LipsonDistilling free-form natural laws from experimental dataScience 324(5923)81ndash85 3 Apr 2009
S Silva and L VanneschiState-of-the-Art Genetic Programming for Predicting Human Oral Bioavailabilityof DrugsIn Proc 4th International Workshop on Practical Applications of ComputationalBiology and Bioinformatics 2010
M SipperLet the Games EvolveIn Proc GPTP 2011
C Sivapragasam R Maheswaran and V VenkateshGenetic programming approach for flood routing in natural channelsHydrological Processes 22(5)623ndash628 2007
C Sivapragasam N Muttil S Muthukumar and V M ArunPrediction of algal blooms using genetic programmingMarine Pollution Bulletin 60(10)1849ndash1855 2010
Bibliographyndash 93
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
L SpectorTowards practical autoconstructive evolution Self-evolution of problem-solvinggenetic programming systemsIn R Riolo T McConaghy and E Vladislavleva editors Genetic ProgrammingTheory and Practice VIII volume 8 of Genetic and Evolutionary Computationchapter 2 pages 17ndash33 Springer Ann Arbor USA 20-22 May 2010
L Spector C Perry and J KleinPush 20 programming language descriptionTechnical report School of Cognitive Science Hampshire College Apr 2004
J Togelius S Karakovskiy J Koutnik and J SchmidhuberSuper Mario EvolutionIn Proc IEEE Computational Intelligence and Games 2009
TORCS The Open Car Racing Simulatorhttptorcssourceforgenet 2012
L Vanneschi and G CuccuVariable Size Population for Dynamic Optimization with Genetic ProgrammingIn Proc GECCO 2009
Bibliographyndash 94
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
E Vladislavleva G Smits and D Den HertogOrder of Nonlinearity as a Complexity Measure for Models Generated by SymbolicRegression via Pareto Genetic ProgrammingIEEE Trans EC 13(2)333ndash349 2009
N Wagner Z Michalewicz M Khouja and R McGregorTime Series Forecasting for Dynamic Environments The DyFor Genetic ProgramModelIEEE Trans EC 2007
J Walker and J MillerPredicting Prime Numbers Using Cartesian Genetic ProgrammingIn Proc EuroGP 2007
J A Walker K Voumllk S L Smith and J F MillerParallel Evolution using Multi-chromosome Cartesian Genetic ProgrammingGPEM 10417ndash445 2009
W-C Wang K-W Chau C-T Cheng and L QiuA comparison of performance of several artificial intelligence methods forforecasting monthly discharge time seriesJournal of Hydrology 374(3-4)294ndash306 2009
Bibliographyndash 95
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-
W Weimer S Forrest C Le Goues and T NguyenAutomatic program repair with evolutionary computationCommunications of the ACM 53(5)109ndash116 June 2010
P Widera J Garibaldi and N KrasnogorGP challenge Evolving energy function for protein structure predictionGPEM 1161ndash88 2010
T YuHierarchical Processing for Evolving Recursive and Modular Programs UsingHigher-Order Functions and Lambda AbstractionGPEM 2345ndash380 2001
Bibliographyndash 96
- Introduction
- Background
- What is genetic programming
- Summary of our first glimpse at GP
- Life demonstration of GP using ECJ
- Assessment of GP techniques
- Case study
- A more detailed view on GP (vanilla GP is not the whole story)
- The Challenges for GP
- Variants of GP
- Selected theoretical results
- Applications of GP
- Additional resources
- Bibliography
-