Bio-Inspired Optimization

Our Journey – For the remainder of the courseA brief review of classical optimization methods

The basics of several stochastic optimization methods evolutionary algorithms (EA) artificial neural networks (ANN) particle swarm optimization (PSO) ant colony optimization (ACO)

The basic biology and physics behind the methods

How to implement and apply the methods

http://www.youtube.com/watch?v=A042J0IDQK4

Optimization:Highest achievable performance by maximizing desired factors

and minimizing undesired onesMaximization: achieving the highest result regardless of cost

or expense $$ Quality of life Etc.

OptimizationIn general, optimization is the

problem of finding the (global) minimum (or maximum) of an objective function

Sometimes (but not always!) the objective function is a specific, well-defined mathematical function

F = f(x1 ,x2, x3 ,...)

Optimization methodsOptimization methods can be divided into two broad

categories Classical (deterministic) optimization methods Stochastic optimization methods

Classical optimization (Covered in ME517) Gradient descent (following the steepest slope) Newton’s method Penalty methods Lagrange multiplier methods

Optimization methodsClassical optimization (Covered in ME517)

Gradient descent (following the steepest slope)http://www.youtube.com/watch?v=HvLJUsEc6dw

Newton’s methodPenalty methodsLagrange multiplier methods

Limitations of classical optimizationClassical methods are less useful in cases with

non-differentiable objective functionsobjective functions whose values can only be obtained as a

result of a (lengthy) simulationVarying number of variables (as in optimization of neural

networks)For such problems, stochastic optimization methods

are more suitable.

Stochastic optimization methodsStochastic optimization algorithms can be used for solving many

problems where classical methods are insufficient More frequently used in industry, particularly in large and complex

problems The number of application areas is steadily increasing

As the name implies, stochastic optimization methods contain an element of stochasticity (randomness)

Many stochastic optimization methods are inspired from biological phenomena

An important subset of stochastic optimization methods are biologically inspired optimization methods.

Biologically inspired optimizationDiscussion

Why would someone use inspiration from biology when defining an optimization method?

What are good examples?

Biologically inspired optimization characteristicsIn comparison with other optimization algorithms, bio-Inspired

optimization algorithms have the following characteristics:The individual components are distributed and autonomous,

there is no central control, and the fault of an individual cannot influence solving the whole problem, these characteristics ensure this kind of algorithms is more robust Is the individual intelligent?

Biologically inspired optimization characteristics

They don’t demand to meet the requirement of differentiability, convexity, and other conditions for mathematical description of the problem This is because of the random search

Use basic mathematical operations, therefore, they are simple and easy to be implemented computationally

These advantages enabled bio-mimetic optimization algorithms to be widely used in a very short period, such as power system, vehicle routing, mechanical design, robotics, etc.

Biologically inspired optimizationNature is all about adaptation, which can be seen as a kind of

optimizationFinding the ideal solution for a specific case (types of eyes)

However, note that in nature the target is constantly moving – unlike the case in engineering problemsMoving surface

Example 1: Shark skinA good example of adaptation (biological optimization)The skin of sharks has evolved to allow low friction very fast

swimming.

Example 1: Shark skinThe rib-like structures affect the interaction between the

surface of the shark and the surrounding water, essentially reducing drag

Prime example of biological inspiration for engineering concepts: Similar ideas are being applied in order to reduce drag (and noise) of aircraft and other vehicles

Shark skin has also inspired the development of the fabric used in swimsuits worn by athletes

Example 2: Evolution of the eyeThe evolution of eyes is another interesting exampleFaced with the problem of generating light-gathering devices,

evolution has come up with no less than 40 completely independent solutions to the problem

Example 3: SwarmingSome species (particularly ants, bees, and termites) display

very advanced forms of cooperationSpecific example: Weaver ants (Oecophylla)

Example 4: SwarmingMany organisms (e.g. some bird and fish species) display

swarming behavior for protection against predators, efficient food gathering, etc...

Biologically inspired optimizationSeveral different optimization methods have been developed,

based on biological phenomena:Evolutionary algorithms (inspired by evolution) Artificial neural networks (inspired by the brain)Ant colony optimization (inspired by cooperative behavior) Particle swarm optimization (inspired by swarming)

EA

Brief introduction to EAsEvolutionary algorithms = EAs.

Based on darwin evolution, which is a process involving gradual, hereditary changes of biological organisms, over long periods of time

Basic biological concepts: Population, fitness, selection, mutation

Individuals that are well adapted to their environment have a high probability of generating offspring

Brief introduction to EAsRandom changes to the

genotype (mutations), sometimes leading to large changes of the phenotype.

Mutations provide new material for evolution to work with

Mutations are random, however selection is not

Basic mode of operation of EAs In an EA, a population of candidate solutions (to the problem at hand) is formed (random

initialization)

Each individual in the population is evaluated and assigned a fitness score based on its performance.

New individuals are formed through the processes of selection, crossover, and mutation.

The new individuals form the second generation, which is evaluated in the same way as the first generation etc.

The process is repeated until a satisfactory solution has been found to the problem at hand.

Evolutionary Algorithms Encoding: the variables of the problem are encoded in strings of digits

known as chromosomes.

The exact encoding procedure varies from problem to problem – in the simple case of optimization of a function f(x1,x2,x3,...xn) the variables xi, i = 1,...,n should be encoded. In more complex problems (e.g. optimization of artificial brains for robots or optimization of neural networks), the specification of the encoding procedure can be much more complicated.

Evolutionary Algorithms Random initialization of a population containing N chromosomes

Evolutionary AlgorithmsCrossover

Mutation

Evolutionary Algorithms

Evolutionary Algorithms

Introduction to Genetic Algorithms http://www.youtube.com/watch?v=zwYV11a__HQ

Optimization Project – More details on wiki 1. Describe the problem.

Identify the specific performance characteristics that are being min/max-imized against each other.

Contrast classical and stochastic optimization, and identify which is more appropriate for your optimization problem.

2. Identify a suitable stochastic optimization algorithm, and explain why you chose it. (You may want to do some outside research.)

3. Solve the optimization problem using a programming language

The fun begins: Boxcar2d.com