Applying natural evolution for solving computational problems - PowerPoint Presentation

342 views
Uploaded On 2022-08-02

Applying natural evolution for solving computational problems - PPT Presentation

First lecture Introduction to Evolutionary Computation Second lecture Genetic Programming Inverted CERN School of Computing 2017 Daniel Lanza CERN Agenda Introduction to Evolutionary Computati ID: 932688

evolutionary selection breeding individuals selection evolutionary individuals breeding evaluation initialization subpop phases fitness species simple computation research tool ecj

Link:

Copy

Embed:

<iframe width="560" height="315" src="https://www.docslides.com/embed/932688" frameborder="0" allowfullscreen></iframe>

Download Presentation from below link

Download Presentation The PPT/PDF document "Applying natural evolution for solving c..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.

Presentation Transcript

Slide1

Applying natural evolution for solving computational problems

First lecture: Introduction to Evolutionary Computation

Second lecture: Genetic Programming

Inverted CERN School of Computing

2017

Daniel Lanza - CERN

Slide2

Agenda

Introduction to Evolutionary Computati

Introduction to Evolutionary AlgorithmsUse cases, practical examplesKey conceptsRepresentation of individualsPhasesEvolutionary Computation research tool (ECJ)

Slide3

Introduction to Evolutionary Algorithms

History

Idea originated

in the 1950sL. Fogel 1962 (San Diego, CA): Evolutionary ProgrammingI. Rechenberg & H.-P. Schwefel

1965 (

Berlin, Germany): Evolution StrategiesWhen to use themWhen finding exact solution is computationally too demanding, but near-optimal solution is sufficient

An evolutionary algorithm (EA) is a heuristic optimization algorithm using techniques inspired by mechanisms from organic evolution such as mutation, recombination, and natural selection to find an optimal configuration for a specific system within specific constraints. [1]

Slide4

Use cases, practical examples

Aircraft wing

design [8][9]

Slide5

Use cases, practical examples

Electronic circuit design, known as evolvable

hardware [2]

Slide6

Use cases, practical examples

Wireless sensor/ad-hoc networks [4].

Slide7

Use cases, practical examples

Vehicle routing problems (traveling salesman problem) [3].

Feature

selection used by machine learning algorithms. [6]Image processing (face recognition) [5].… [7]

Slide8

Key concepts

A population of individuals is evolved through generations

Each individual’s genome describes a candidate solution

The fitness function evaluates new individualsThe evolutionary process is finished as soon as the optimal solution is found

Slide9

Representation of individuals

It should be able to represent all the search space

But should not able to represent impossible solutions

Examples of solutions to different problems:[1, 0, 1, 1, 0, 1, 1]

[4.5, 1, 100.3, 9, 21, 934, 1]

[“right”, “left”, “up”, ”up”, “left”, “left”, “down”]

[4.5, “left”, false, true, 9]

Slide10

Representation of individuals: example

Choose points of interest for face recognition [5]

Slide11

Phases

Similarly to how natural evolution works

…

Initialization

Randomly generated

Evaluation

Calculate fitness for each individual

BreedingIndividuals are crossed-over and mutations take place

Selection

Choose individuals for breeding

Optimal solution

Slide12

“MaxOne

” problem

Starting from randomly generated strings of 0s and 1s

Evolve to the optimal solution, a string of 1sPhases: in detail by using an example[1, 1, 1, 1, 1, 1, 1]

Slide13

The first population is filled up with individuals

The individuals

are randomly generated following problem’s criteria

The population size is determined beforehand and remains fixedPhases: initialization

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[0,

, 1, 1, 0, 1, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 1, 1, 1, 0, 0,

]

[0, 0, 0, 1, 0, 0, 1]

Slide14

Phases: evaluation

Practically the fitness function defines the problem

Get the fitness of each individual

Fitness describes how well the individual solves the problemIn “MaxOne”, fitness is defined as the number of 1s in the individual

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[0,

, 1, 1, 0, 1, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 1, 1, 1, 0, 0,

]

[0, 0, 0, 1, 0, 0, 1]

Slide15

Phases: selection

A selection strategy is defined to choose the parents

Selected individuals, parents, will be used for breeding

Fitness is taken into account (best individuals)But also some randomness affects the selection (simulating real life)Individuals with reduced fitness could have valuable featuresElitism (optional): best individual is copied to the next generation

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[0,

, 1, 1, 0, 1, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 1, 1, 1, 0, 0,

]

[0, 0, 0, 1, 0, 0, 1]

Slide16

Phases: selection (techniques)

Tournament [11]

A fixed number of individuals are randomly selected

Among them, the best one is selected

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[0,

, 1, 1, 0, 1, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 1, 1, 1, 0, 0,

]

[0, 0, 0, 1, 0, 0, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 1, 1, 1, 0, 0,

]

Slide17

Phases: selection (techniques)

Roulette-wheel

[12]

Selection probability is proportional to individual’s fitnessOthers… [13]

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[0,

, 1, 1, 0, 1, 1]

[0, 0, 1, 1, 0, 0, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 1, 1, 1, 0, 0,

]

[0, 0, 0, 1, 0, 0, 1]

Slide18

Phases: breeding

Selected individuals cross-over

New individuals fill up the next generation of individuals

Selection and breeding phases are performed till next population is filled

Initialization

Evaluation

Breeding

Selection

[1, 0, 1, 1, 0, 1, 1]

[1, 0, 0, 1, 1, 1, 1]

[1, 0, 0, 1, 0, 1, 1]

[1, 0, 1, 1, 1, 1, 1]

Parents

Offspring

Slide19

Phases: breeding (techniques)

Some times features cannot be mixed

Smarter operations need to be applied

For example:Depends on the problemNumber: average, max, min, sum, ..Boolean: and, or, xor, …Strings: concatenate, replace, remove, split,

…

Initialization

Evaluation

BreedingSelection

[4.5, “left”, false, true, 9]

[1.2, “right”, true, true, 9]

Parents

Slide20

Phases: mutation

With a very little likelihood

A random modification is applied

Initialization

Evaluation

Breeding

Selection

[1, 0, 0,

, 0, 1, 1]

[1, 0, 0,

, 0, 1, 1]

Slide21

Phases: mutation (techniques)

Any modification can be consider as a mutation

More complex genomes could have their features modified to any possible value

Initialization

Evaluation

Breeding

Selection

[1, 0, 0,

, 1, 1]

[1, 0, 0,

, 1, 1]

[

1, 0, 0, 1, 0, 1, 1

]

[

1, 1, 0, 1,

, 1

]

[4.5, “left”, false, true, 9]

[4.5,

“right”

, false, true, 9]

[

0.5

, “right”, false, true, 9]

[4.5, “right”, false,

false

, 9]

Slide22

Phases: evaluation,

selection, breeding

…The loop keep going till the optimal individual is found

Initialization

Randomly generated

Evaluation

Calculate fitness for each individualBreeding

Individuals are crossed-over and mutations take place

Selection

Choose individuals for breeding

[1, 1,

, 1, 1,

, 1]

Slide23

Evolutionary Computation research tool (ECJ)

Developed at the George Manson University [10]

Eliminates the

need of implementing the

evolutionary

process

Highly used in the communityMain features:

Multi-platform: JavaFlexibility: easy to implement many kind of problemsConfiguration filesCheck pointsMulti-threadingPseudo-random number generator: reproduce results23

Slide24

Evolutionary Computation research tool (ECJ)

Code architecture allows pluggable and customized components

Several built-in implementations for every component

Slide25

Evolutionary Computation research tool (ECJ)

Configuring the

MaxOne

problem

breedthreads

= 1evalthreads = 1seed.0 = 4357state = ec.simple.SimpleEvolutionStatepop = ec.Population

init = ec.simple.SimpleInitializerfinish = ec.simple.SimpleFinisherbreed = ec.simple.SimpleBreedereval = ec.simple.SimpleEvaluatorstat = ec.simple.SimpleStatisticsexch = ec.simple.SimpleExchangergenerations = 200pop.subpops = 1pop.subpop.0 = ec.Subpopulationpop.subpop.0.size = 10pop.subpop.0.species = ec.vector.BitVectorSpeciespop.subpop.0.species.fitness = ec.simple.SimpleFitnesspop.subpop.0.species.ind = ec.vector.BitVectorIndividualpop.subpop.0.species.genome-size = 20pop.subpop.0.species.mutation-type = flippop.subpop.0.species.mutation-prob = 0.01pop.subpop.0.species.pipe = ec.vector.breed.VectorMutationPipelinepop.subpop.0.species.pipe.source.0 = ec.vector.breed.VectorCrossoverPipelinepop.subpop.0.species.pipe.source.0.source.0 = ec.select.TournamentSelectionpop.subpop.0.species.pipe.source.0.source.1 = ec.select.TournamentSelectioneval.problem = ec.app.tutorial1.MaxOnes

Slide26

Evolutionary Computation research tool (ECJ)

Implementing the

MaxOne

fitness function: ec.app.tutorial1.MaxOnes

Slide27

Evolutionary Computation research tool (ECJ)

Execution

Slide28

Questions?

Slide29

Applying natural evolution for solving computational problems

First lecture: Introduction to Evolutionary Computation

Second lecture: Genetic Programming

Tomorrow at 11:30

Slide30

References

[

] https://en.wikipedia.org/wiki/Evolutionary_algorithm[2] W. Greenwood, Garrison; Tyrrell, Andrew M. (2006-10-20). Introduction to Evolvable Hardware: A Practical Guide for Designing Self-Adaptive Systems (1 ed.). Wiley-IEEE Press. ISBN 978-0471719779

[3]

Maimon, Oded; Braha, Dan (1998). "A genetic algorithm approach to scheduling PCBs on a single machine" (PDF). International Journal of Production Research

.[4] BiSNET/e – Distributed Software Systems Group, University of Massachusetts, Boston[5] D. Lanza, F. Chavez, F. Fernandez, C. Benavides-Alvarez and J. Villegase, Speeding up Evolutionary Approaches to Face Recognition by Means of Hadoop. EVO 2016[6] Haleh Vafaie and Kenneth De Jong. Genetic Algorithms as a Tool for Feature Selection in Machine Learning. Center for Artificial Intelligence, George Mason University[7] https://en.wikipedia.org/wiki/List_of_genetic_algorithm_applications#cite_note-56[8] Andre C. Marta. Parametric Study of a Genetic Algorithm using a Aircraft Design Optimization Problem. Stanford University, U.S.A.[9] I.

Kroo

Aeronautical

Applications

Evolutionary

Design.

Stanford University, U.S.A

[10]

https://cs.gmu.edu/~eclab/projects/ecj

Slide31

References

[11] Miller

, Brad; Goldberg, David (1995). "Genetic Algorithms, Tournament Selection, and the Effects of Noise".