Case studies in applying fitness distributions in evolutionary algorithms: I. Simple neural networks and Gaussion mutation

摘要

Abstract: Evolutionary algorithms are often applied to taskswhere the challenges is to find a superior solution.The engineering challenge concerns how to best designsuch algorithms in terms of their representation,variation operators, and selection. The distribution offitness scores that is obtained when applying variationoperators to parents can provide useful information forsetting the parameters that are associated with thoseoperators. Experiments presented here indicate thatfitness distributions can also reveal information aboutthe landscapes that surrounds particular parents andsuggest that typical methods of self-adaption may notbe very well suited for exploring the state space ofpossible solutions in the presence of multiple minima.!20