Slide 1

FSS (F ISH S CHOOL S EARCH ) I NTELL. A LGORITHMS FOR O PTIMIZATION Prof. Carmelo Bastos Filho, PhD Prof. Fernando Buarque de Lima Neto, DIC PhD Computational Intelligence Research Group (CIRG) Pernambuco Polytechnic School of Engineering (POLI) University of Pernambuco (UPE) Recife, Brazil. X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence S HORT -C OURSE 5 Slide 2 CIRG@UPE Recife 2 Computational intelligence research group 5 Professors 14 M.Sc. students 17 B.Sc. students Slide 3 Agenda Part 1: Fundamentals -Swarm Intelligence -FSS: motivation and inspirations -FSS: vanilla operators Part 2: Developments -Multimodal Optimization -FSS: versions d and p -FSS: applications and results -FSS: the way ahead X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 3 Prof. Carmelo Bastos Filho, PhD carmelofilho@poli.br Prof. Fernando Buarque de Lima Neto, DIC PhD fbln@ecomp.poli.br Slide 4 P ART - 1: FUNDAMENTALS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 4 Slide 5 P ART - 1: FUNDAMENTALS SWARM INTELLIGENCE X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 5 Slide 6 Swarm Intelligence Simple entities Emergence of intelligence to solve complex problems Examples PSO ACO ABC BFA X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 6 Slide 7 Particle Swarm Optimization (PSO) Proposed in 1995 by Kennedy and Eberhart Attributes: Current position in the search space x(t) Current velocity v(t) Best position found by the particle during the search (pbest) Best position found by the particles neighbors during the search (gbest) Velocity update equation V(t+1) = w.v(t) + c 1.r 1.[p best -x(t)] + c 1.r 1.[g best -x(t)] Position update equation x(t+1) = x(t) + v(t) Social term Cognitive term X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 7 Slide 8 PSO Fast convergence, but quickly looses diversity X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 8 Slide 9 P ART - 1: FUNDAMENTALS FSS: MOTIVATION AND INSPIRATIONS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 9 Slide 10 In real-world problems, search spaces can be huge Computationally intensive techniques may not scale-up properly or tackle a sizeable number of constraints Gregarious fish are much better equipped to survive (finding food in massive volumes and defending themselves from various threats) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 10 Slide 11 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 11 Slide 12 FSS was invented to be a new* family of algorithms (within Swarm intelligence approach, thus particle based) suited for optimization in high-dimensional search spaces (and now able to split the school). FSS was devised so that all fish perform local search while the fish school aggregates social information * Invented by Bastos Filho & Lima Neto in 2008 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 12 Slide 13 Recife 2007... X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 13 Slide 14 First publication by Bastos Filho & Lima Neto in 2008 http://www.fbln.pro.br/FSS/people.htm X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 14 Slide 15 ICSI 2011 Chongqing, China X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 15 Slide 16 What is fss about ? TERMINOLOGY & RATIONALE X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 16 Slide 17 Fishes are entities of the swarm (i.e. school) Aquarium is the search space that can be of high dimensionality Position of each fish within the aquarium is one candidate solution (i.e. a set of values for the parameter vector) of the optimization process Weight of each fish indicates its individual success (i.e. fitness) in finding good solution Radius of the fish school indicates the collective success in finding good solution X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 17 Slide 18 (1)'Swimming' actually is a means of: performing a local search + Storing information of success (Both of Individuals and Collective nature) (2)Success of the search is given by: fish weights (large is better) + school radius (small is better) + School barycenter (closer to optima is better) (3)Non-monotonicity is achieved, e.g.: (i) By random hesitation before swim + (ii) By expansion/shrinking the school radius + (iii) By variations on swimming components X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 18 Slide 19 What is FSS about ? FUNCTIONING & OPERATORS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 19 Slide 20 Local computations (i.e. swimming) is composed of distinct components (this results in smart swimming) Fish think they are alone in the aquarium, but whenever they need there is an intuition available (this results in very low communication cost) Barycenter of the school moves smoothly towards optimal solution (this results in always better solutions along processing) School radius is based on barycenter gradient (this results in the interesting ability to self control exploration and exploitation modes of operation) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 20 Slide 21 P ART - 1: FUNDAMENTALS FSS: VANILLA OPERATORS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 21 Slide 22 OPERATORS: 1. Feeding 2. Swimming: Individual movement Collective-instinctive Collective-volitive STOP-CONDITIONS: 1. limit of cycles; 2. time limit; 3. maximum school weight 4. minimum school radius (*) Learning Local search Social glue Global search Problem dependent Problem independent (#2) (#1) (#3) (#4) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 22 Slide 23 FSS Operators: Individual movement Feeding Collective-instinctive movement Collective-volitive movement X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 23 Slide 24 Fish Swimming - Individual X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 24 Slide 25 FSS Operators: Individual movement Feeding Collective-instinctive movement Collective-volitive movement X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 25 Slide 26 FSS Operators: Individual movement Feeding Collective-instinctive movement Collective-volitive movement X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 26 Slide 27 Fish Swimming Collective instinctive X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 27 Slide 28 FSS Operators: Individual movement Feeding Collective-instinctive movement Collective-volitive movement This operator performs contraction or expansion depending on the school of fish success or failure - Barycenter calculation: - Volitive movement Expansion (+) Contraction (-) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 28 Slide 29 Fish Swimming Collective vollitive X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 29 Slide 30 Single-objective optimization (SOO) Static Environment Examples of FSS at work Sphere Function (3D view) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 30 Slide 31 FSS Is capable to auto-regulate the granularity of search, but has slow convergence X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 31 Slide 32 Single-objective optimization (SOO) Static Environment Examples of FSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 32 Slide 33 P ART - 2: DEVELOPMENTS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 33 Slide 34 P ART - 2: DEVELOPMENTS M ULTIMODAL O PTIMIZATION X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 34 Slide 35 The world is not flat... Pico da Neblina Brazilian highest summit @ 3,040m A-S-L In NW Brazilian Amazon Forest, close to the border with Venezuela X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 35 Slide 36 Sometimes it seems to be flat... Monte Roraima Impressive 30 Km 2 summit @ 2,772m A-S-L In Guyana, close to the triple border of Brazil and Venezuela X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 36 Slide 37 BUT it is not flat at all! Mont Serrat Impressive set of peaks @ 1,236m A-S-L In Catalonia, 50Km far from Barcelona in NE of Spain X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 37 Slide 38 How FSS could be improve from tackling only one global solution to: (i) many/all global solutions or even (ii) many/local solutions of a given MMOP? (i) Mount-Roraima MMOPs (ii) Mountserrat MMOPs (Infinite number of solutions) (Finite number of solutions) Evokes impossibility Evoke unfeasibility X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 38 Slide 39 P ART - 2: DEVELOPMENTS FSS: VERSION D X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 39 Slide 40 Fish schools do split in Nature! X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 40 Slide 41 Large fish schools are good (for protection and for finding food) but are also bad (because of increased competition) The splitting of the school could be governed by the density of fish within a sub-regions of the school Densely populated regions would provide food to closer fish much more efficiently than to fish that are apart (a sort of collaborative behavior) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 41 Slide 42 The new flavor of Fish School Search, dFSS, was devised to be a metaheuristic that is non-exhaustive, fast, scalable, inexpensive and is able to elegantly tackle multimodal optimization problems. X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 42 Slide 43 Fishes are entities of the swarm (i.e. school) Aquarium is the search space that can be of high dimensionality Position of each fish within the aquarium is one candidate solution (i.e. a set of values for the parameter vector) of the optimization process Weight of each fish indicates its individual success (i.e. fitness) in finding good solution Radius of the fish school indicates the collective success in finding good solution Same as FSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 43 Slide 44 (1)'Swimming' actually is a means of: performing a local search + Storing information of success (Both of Individuals and Collective nature) (2)Success of the search is given by: fish weights (large is better) + school radius (small is better) + School barycenter (closer to optima is better) (3)Non-monotonicity is achieved, e.g.: (i) By random hesitation before swim + (ii) By expansion/shrinking the school radius + (iii) By variations on swimming components Almost the same as FSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 44 Slide 45 Every fish share their food with all others Sharing of food depends on (i) distance and (ii) density Distance and density act as segregators (among sub- populations of the school) Fish now has a memory of collaborations with their pals Swimming in dFSS is weighted also by this memory X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 45 Slide 46 OPERATORS: 1. Feeding 2. Swimming: Individual movement Collective-instinctive Collective-volitive STOP-CONDITIONS: 1. limit of cycles; 2. time limit; 3. maximum school weight 4. minimum school radius (*) Learning Local search Social glue Global search Problem dependent Problem independent (#2) (#1) (#3) (#4) Almost the same as FSS (two more: memory and partition) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 46 Slide 47 Individual movement FSS dFSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 47 Slide 48 Feeding FSS dFSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 48 Slide 49 MEMORY (*NEW*) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 49 Slide 50 Collective-instinctive FSS dFSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 50 Slide 51 Collective-VOLITIVE FSS dFSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 51 Slide 52 PARTITION (*NEW*) while There is fish in the main school do Choose a fish i randomly in the main school Create a new subgroup Si Put fish i in subgroup Si Remove fish i from the main school Find other fish j in the main school that satisfies (#) while there exists fish j in the main school do Put fish j in subgroup Si Remove fish j from the main school Set i = j Find other fish j in the main school that satisfies (#) end while X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 52 Slide 53 Multi-modal optimization (MMOP) Equal-Peaks A Function Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 53 Slide 54 Multi-modal optimization (MMOP) F.O.M. Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 54 Slide 55 Comparing FSS and dFSS (Uni- and Multimodal abilities) Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 55 Slide 56 Multi-modal optimization (MMOP) 3D-Function: Plateus Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 56 Slide 57 Multi-modal optimization (MMOP) 3D-Function: Staircase Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 57 Slide 58 Multi-modal optimization (MMOP) 3D-Functions: Circles Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 58 Slide 59 Benchmark functions used for comparison Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 59 Slide 60 Comparing dFSS to GSO and NichePSO (Check ICSI2011 Paper MADEIRO et al.) Metric: average on 30 trial only when algorithms were able to capture above 95% of existing optimal solutions of the MOOP Examples of dFSS at work X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 60 Slide 61 P ART - 2: DEVELOPMENTS FSS: VERSIONS P X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 61 Slide 62 Fish (in schools) do process stand alone (sometimes)! X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 62 Slide 63 Why not using graphic processors to boost FSS computation? CUDA Platform (NVIDIA) is easy to be used for that! X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 63 -GeForce GTX 280 (GPU compatible with NVIDIA CUDA ) -1296MHz (240 Cores) -1Gb Memory -CUDA: 3.2 -OpenGL: 2.1 -Operating System: Ubuntu 10.04 Slide 64 The new flavor of Fish School Search, pFSS, was devised to be a metaheuristic that is non-exhaustive, fast, scalable, inexpensive and in a parallel manner. X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 64 Slide 65 Fishes are entities of the swarm (i.e. school) Aquarium is the search space that can be of high dimensionality Position of each fish within the aquarium is one candidate solution (i.e. a set of values for the parameter vector) of the optimization process Weight of each fish indicates its individual success (i.e. fitness) in finding good solution Radius of the fish school indicates the collective success in finding good solution Same as FSS X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 65 Slide 66 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 66 Synchronous Asynchronous Slide 67 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 67 Simulation details: -30 fish -30 dimensions -50 executions (average) -10.000 iteracion Rosenbrock Rastrigin Griewank Slide 68 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 68 Slide 69 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 69 CIRGs newest PSC (with 1792 cores) Slide 70 P ART - 2: DEVELOPMENTS FSS: THE WAY AHEAD X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 70 Slide 71 FSS: The way ahead Hybridization with other Swarm Techniques* Parallelization of current implementations** New operators** Use in Dynamic Problems* Speciation** New real-world complex applications*** moFSS (Multi-Objetive)*** Legend: * Results already published ** Research already initiated *** Some possibilities already assessed X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 71 Slide 72 FSS: The way ahead X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 72 FSS is fast (and outperforms many other Swarm Intelligent algorithms) FSS is computationally inexpensive (almost no communication costs) FSS controls its own exploration/exploitation mode along search FSS is flexible for new extensions and modifications dFSS deals with multimodal problems elegantly New investigations indicate that FSS is easy to parallelize and can tackle dynamic problems Slide 73 X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 73 ICSI 2012 Shenzhen, China *NEW Slide 74 O FFICIAL WEB - SITE X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 74 Slide 75 M AIN P AGE (W EB ) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 75 HTTP :// WWW. FBLN. PRO. BR /FSS/ INDEX. HTM Slide 76 L INKS P AGE (W EB ) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 76 HTTP :// WWW. FBLN. PRO. BR /FSS/ LINKS. HTM Slide 77 V ERSIONS P AGE (W EB ) X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 77 HTTP :// WWW. FBLN. PRO. BR /FSS/ VERSIONS. HTM Slide 78 REFERENCES X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 78 Slide 79 References MADEIRO, Salomo S.; BASTOS-FILHO, Carmelo J. A.; LIMA NETO, Fernando B. de. "Multimodal Optimization based on Fish School Behavior" to appear in Swarm Intelligence Journal, 2010, Springer. [TO APPEAR] CAVALCANTI JNIOR, G. M. ; Bastos-Filho, C. J. A. ; LIMA NETO, F. B. ; CASTRO, R.. A Hybrid Algorithm based on Fish School Search and Particle Swarm Optimization for Dynamic Problems. In ICSI2011: Second International Conference on Swarm Intelligence. Springer - Lecture Notes in Computer Science, v. 6729, p. 543-552, 2011. MADEIRO, S. S. ; LIMA NETO, F. B. ; Bastos-Filho, C. J. A. ; FIGUEIREDO, E. M. N.. Density as the Segregation Mechanism in Fish School Search for Multimodal Optimization Problems. In ICSI2011: Second International Conference on Swarm Intelligence. Springer - Lecture Notes in Computer Science, v. 6729, p. 563-572, 2011. BASTOS-FILHO, Carmelo J. A.; LIMA NETO, Fernando B. de; SOUSA, Maria F. C.; PONTES, Murilo R.; MADEIRO, Salomo S. "On the Influence of the Swimming Operators in the Fish School Search Algorithm". In: IEEE International Conference on Systems, Man, and Cybernetics - SMC2009, 2009, San Antonio, USA. BASTOS FILHO, Carmelo J. A. ; LIMA NETO, Fernando B. de; LINS, Anthony J. C. C.; NASCIMENTO, Antnio I. S.; LIMA, Marlia P. "A Novel Search Algorithm based on Fish School Behavior". In: 2008 IEEE International Conference on Systems, Man, and Cybernetics - SMC 2008, 2008, Cingapura. BASTOS-FILHO, Carmelo J. A.; LIMA NETO, Fernando B. de; LINS, Anthony J. C. C.; NASCIMENTO, Antnio I. S.; LIMA, Marlia P. "Fish School Search: an overview". In: CHIONG, Raymond (Ed.). Nature-Inspired Algorithms for Optimisation. Series: Studies in Computational Intelligence, Vol. 193.. pp. 261-277. Berlin: Springer- Verlag, 2009. {ISBN: 978-3-642-00266-3} X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence FSS Intelligent Algorithms for Optimization Profs. Carmelo Bastos Fo. e Fernando Buarque (CIRG/UPE) Fortaleza-CE, 08/11/2011 Brazil Slide 79 Slide 80 Prof. Carmelo Bastos Filho, PhD carmelofilho@poli.br Prof. Fernando Buarque de Lima Neto, DIC PhD fbln@ecomp.poli.br Computational Intelligence Research Group (CIRG) Pernambuco Polytechnic School of Engineering (POLI) University of Pernambuco (UPE) Recife, Brazil. X Brazilian Congress on Computational Intelligence XI School on Computational Intelligence V ALEU MACHO ! (T HAT IS, THANK YOU IN THE LOCAL DIALECT ) Follow the fish...