Transcript
Page 1: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Monte Carlo Methods and the Genetic Algorithm

Definitions and Considerations

John E. NawnMAT 5900March 17th, 2011

Page 2: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

What is the Genetic Algorithm?

Heuristic search method employing randomness in order to determine the optimal solution to a wide range of problems

Applications include:◦Economics◦Number Theory◦Rankings◦Path Length Determination (TSP, etc.)

Based in Neo-Darwinian theory

Page 3: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

History of Genetic AlgorithmsOperational Research (1940s and

1950s) – birth of heuristicsEvolutionsstrategie – Rechenberg

and Schwefel (1960s)Adaptation in Natural and

Artificial Systems – John Holland (1975)

Increased computational complexity (1990s – 2000s)

Page 4: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Evolution: A SurveyOn the Origin of Species – Charles

Darwin (1859)Proposed natural selection –

environment creates selection pressure for individuals in a species

Selected advantages may be heritable: provides method for determining fitness of offspring

What Darwin (and biologists) didn’t know…

Page 5: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Genetics: A SurveyGregor Mendel (1863)Individuals within a species carry

directions for their promulgationSegregation (First Law)Independent Assortment (Second

Law)Increasing technology and the

discovery of mutations and crossovers

Genotype and phenotype

Page 6: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

TerminologyPopulation

◦Set of possible solutions in any given generation

Chromosomes◦Basic units that undergo reproduction

in the algorithm◦Two types: binary and non-binary◦Minimum size requirements◦Genes and alleles

Reproduction

Page 7: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Terminology Mutation

◦Process of changing allele values in a chromosome

◦Inversions◦How often?◦What type?

Crossover◦Process of combining parental

chromosomes to yield new chromosomes

◦What type?

Page 8: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

TerminologySelection

◦Criterion◦Fitness functions◦Reeves and Rowe:

Tournament selection Ranking

Termination◦Diversity thresholds◦Generation limits◦Computational limits

Page 9: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Minimum String Length Requirements

Reeves, Colin R.; p. 28

Page 10: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

MutationsSimplicity of methodBinary

◦Reversal of allelesNon-binary

◦Stochastic selection of new alleles◦Differing mutation rates◦Selecting complete mutations and

error repair

Page 11: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Crossovers (X)Binary

◦NX – N-point crossovers◦UX – Uniform crossover, or linear

operator “masks” Non-Binary

◦Difficulty in applying n-point crossovers◦PMX – Partially matched crossover◦UX – “in/out” order crossovers

Further possibilities – Fox/ McMahon and Poon/ Carter

Page 12: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Fitness FunctionsMethod comparing gene successRoulette wheel model of selectionSelection pressure =

individual fitness/ total fitnessBenefit of larger selection

pressureNiches

Page 13: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Critiques of the Genetic Algorithm:Biological and Philosophical ArgumentsWhat is natural selection

selecting for?Evolution as a theory or fact: Lisa

GatlinIndividual genes and group

interactions Lamarckian or Darwinian

evolution?

Page 14: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Critiques of the Genetic Algorithm:Mathematical ArgumentsLack of theory in heuristic

applicationsNewton’s Method problemBest possible solution or best

solution?Pseudo-randomnessSimilarities to Markov chains and

processes (a.k.a. t – 1 dependency)

Page 15: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

What to Expect NextCrossover possibilitiesHolland’s method - schemata

approachesThree applications:

◦General Path Problems or the Traveling Salesman Problem (TSP)

◦Ranking Styles◦Stock Selection

Page 16: Monte Carlo Methods and  the Genetic Algorithm Definitions and Considerations

Selected BibliographyCraig, Nancy L. et. al. Molecular Biology:

Principles of Genome Function. New York: Oxford University Press, 2010. Print.

Krzanowski, Roman and Jonathan Raper. Spatial Evolutionary Modeling. New York: Oxford University, Inc., 2001. Print.

Reeves, Colin R. and Johathan E. Rowe. Genetic Algorithms: Principles and Perspectives: A

Guide to GA Theory. Boston: Kluwer Academic Publishers, 2003. Print.

Russell, Peter J. iGenetics: A Mendelian Approach. San Francisco: Pearson Education, Inc., 2005. Print


Recommended