Aggregating Models of Evolutionary Algorithms

This paper summarizes two useful techniques for aggregating theoretical models of evolutionary algorithms (EAs). Aggregation removes unnecessary detail from the models, producing simpler models with good predictive accuracy. The first aggregation technique is applicable to “equation of motion” models of EAs that have selection and mutation, for a particular class of interesting fitness functions. This form of aggregation introduces no error into the theoretical model. The second aggregation technique is more general – it can be applied to arbitrary Markov chain models of dynamic systems, such as EAs with selection, mutation, and recombination. No assumptions are made about the fitness functions. This form of aggregation introduces only a small amount of error into the theoretical model.