Lecture 5 - Evolutionary Algorithms I Flashcards
What is an Evolutionary Algorithm?
Evolutionary algorithms simulate natural evolution of individual structures at the genetic level using the idea of survival of the fittest via processes of selection, mutation, and reproduction.
An individual represents a candidate solution
A collection of individuals currently ‘alive’, called population is evolved from one generation to another (iterations) depending on the fitness of individuals in a given environment, indicating how fit an individual is (how close to the optimal solution)
What are the basic components of a GA (Genetic Algorithm)?
A genetic representation for candidate solutions
An initialisation scheme to generate the first population of solutions
A fitness function that plays the role of environment, rating solutions in terms of their fitness.
A scheme for selecting mates (parents) for recombination
Crossover (recombination) exchanges genetic material between mates producing offspring (children)
Mutation perturbs an individual creating a new one
Replacement strategy to select the surviving individuals for the next generation
Termination criteria
Values for various parameters that GA uses (population size, probabilities of applying genetic operators, etc…)
What are the features of GA representation?
Each individual contains one chromosome, and has an associated fitness value
Chromosomes contain a fixed number of genes - chromosome length
Traditionally, binary encoding is used
What are the features of GA initialisation?
Population size - number of individuals are created randomly (random initialisation)
Each gene at a locus of an individual is assigned an allele value of 0 or 1 randomly, decided by flipping a coin
Length of chromosomes is determined by the amount of different variables in the MAX-SAT problem i.e. length of 6 if variables = {a,b,c,d,e,f}
What are the features of GA fitness calculation?
Fitness value indicates how fit the individual is to survive and reproduce under the current conditions, and how much the current solution meets the requirements of the objective function
Fitness value is obtained by applying the fitness function to the individual’s chromosome
What are the features of GA reproduction?
Reproduction consists of:
Selecting individuals - apply selection pressure considering the fitness of individuals in the population e.g. Roulette Wheel Selection
Usually 2 parents are selected using the same method, which will go under the crossover operation.
What is Roulette Wheel Selection?
Fitness level is used to associate a probability of selection with each individual chromosome
While candidate solutions with a higher fitness will be less likely to be eliminated, there is still a chance that they may be
Expected number of representatives of each individual in the pool is proportional to its fitness.
What is Tournament Selection?
This method involves running a number of ‘tournaments’ among randomly chosen individuals (of tour size) selecting the one with best fitness at the end - this process is repeated for selecting each parent to be recombined
What are the features of recombination - crossover?
Selected pairs are recombined to form new individuals - this is exchange of genetic material
Crossover is applied with a crossover probability Pc, which in general is chosen close to 1.0
What are the different types of crossover (GA)?
One point crossover - generate a random number in [0,1], if it is smaller than a crossover probability, then select a random crossover site in [1, chromosome length], split individuals at the selected site, and then exchange segments between pairs to form two new individuals
2 Point crossover - same as above, except selects two different points, thus splitting it three ways, and then swaps them around
Uniform Crossover - Goes through each bit of the parent string, and exchanges the two if the random number between [0,1] is greater than 0.5.
What is mutation within GAs?
Loop through all the alleles of all the individuals one by one, and if that allele is selected for mutation with a given probability Pm, you can either change it by a small amount or replace it with a new value - for binary representation, this is actually just flipping a bit
Mutation rate is typically very small (0.001). Choosing Pm as (1/chromosome length) implies on average a single gene will be mutated for an individual.
What features of replacement are there for GAs?
Generation gap (alpha) controls the fraction of the population to be replaced in each generation, where alpha is a number between [1/N, 1.0] - number of offspring produced at each generation is g = alpha*N
What is Steady-State GA replacement?
g=2, that is alpha = 2/N
Two offspring replace two individuals from the old generation, choosing one of four methods:
Two offspring replace two parents
Two offspring replace worst two of population
Best two of parents and offspring replace two parents (called elitism)
Best two of parents and offspring replace worst two of the population (strong elitism)
What are the features of termination criteria in GAs?
The evolution continues until a termination criteria is met, possibly until:
A predefined number of generations has been exceeded
A goal is reached
Best fitness does not change for a while
A condition is satisfied depending on a combination of above
What is convergence?
Defined as the progression towards uniformity (individuals become alike):
Gene convergence - a location on a chromosome is converged when 95% of the individuals have the same gene value for that location
Population convergence - a population is converged when all the genes have converged
Phenotypic convergence - average fitness of the population approaches to the best individual in the population
