Evolutionary Algorithms: Exploring the Dynamics of Self-Adaptation

Abstract

Self-adaptation refers to allowing characteristics of search--most often mutation rates--to evolve on a per-individual basis rather than be specified by the user. This practice is gaining increasing attention and moving beyond classical mutation rates to explore other traits affecting search. The potential impact of self-adaptation is vast because it provides an implicit approach to problems of operator selection and parameter tuning, and possibly to those of representation as well. Studies have demonstrated many successful applications of self-adaptation, but in light of its potential impact, it is important to gain insight into the dynamics of this process to guide further experimentation. To this end, we present here an illuminating relationship between the strength of selection pressure and the magnitude of self-adapting mutation rates, as well as an observation on when self-adapting mutation rates are most likely to be of greatest utility.