Evolutionary Computation

1
Evolutionary Computation Evolution Strategies

• Winter 2005

2
Biology 101

• Charles Darwin Origin of Species
• Life through evolution
• Gradual changes to beings over time
• Evolution powered by natural selection
• Individuals best adapted to their environment
  have a higher chance of survival
• i.e. surival of the fittest
• Survival passing ones genes to the new
  generation

3
Genetics 101

• Organism blueprint stored in DNA
• DNA information storage molecule
• Complete DNA stored in all cells
• Double helix structure
• 4 building blocks for information storage
• Adenosine, Thyamine, Guanine, Cytosine
• Pairing of bases between 2 comlementary strands
• A and T, G and C
• DNA structured into chromosomes

4
Genetics 101 part 2

• Chromosomes divided into genes
• Genes code for protein
• Genes determine traits of an individuals
• i.e. eye color, hair color, etc.

5
Genetics 102

• Genotype
• Genetic makeup of an organism
• i.e. the DNA information
• e.g. coding of the gene for eye-color
• Phenotype
• Appearance or characteristics of an organism
  arising from interactions of its genetic makeup
  with the environment
• e.g. blue eye color

6
Genetics 102 part 2

• Genetic Operators
• Reproduction (asexual)
• copies of individual genetic code form a new
  individual
• Recombination (crossover/sexual reprod.)
• New individual arises from a mixture of the
  genetic code of the 2 parents
• Mutation
• Small changes in genetic code due to copying
  errors or environmental factors

7
Evolutionary Computation

• Use the ideas stemming from natural evolution in
  order to solve problems
• Algorithms mainly search algorithms
• Various approaches we will look at
• Evolution Strategies (ES)
• Genetic Algorithms (GA)
• Genetic Programming (GP)

8
Evolution Strategies (ES)

• Rechenberg Schwefel (Europe)
• From Engineering perspective
• Simple chromosome encoding
• e.g. real numbers, ASCII codes
• Genetic operators
• mainly mutation
• can also use recombination
• Idea of evolving strategy parameters

9
Simple ES Algorithm

• Generate initial set of n individuals
• S s1, , sn
• Evaluate all individuals to calculate fitness
  using a fitness measure
• Select the best individual sbest?S
• Generate n-1 mutants from sbest
• M si mut(sbest) i1,,n-1
• Evaluate mutants to calculate fitness
• Stop if maximum fitness reached or set Ssbest
  U M and continue with step 3