Population aggregation based on fitness

Abstract

It is known that modelling a finite population genetic algorithm as a Markov chain requires a prohibitively large number of states. In an attempt to resolve this problem, a number of state aggregation techniques have been proposed. We consider two different strategies for aggregating populations, one using equal average fitness and the other using equal best fitness. We examine how the approximation scales with population size, in addition to studying the effects of other parameters (such as mutation rate). We find that a large reduction in the number of states is possible, sometimes with surprisingly small loss of accuracy.