Spatially-Structured Sharing Technique for Multimodal Problems

Spatially-structuredpopulations are one approach to increasing genetic diversity in anevolutionary algorithm (EA). However, they are susceptible toconvergence to a single peak in a multimodal fitness landscape.Niching methods, such as fitness sharing, allow an EA to maintainmultiple solutions in a single population, however they have rarelybeen used in conjunction with spatially-structured populations.This paper introduces \it local sharing, a method that appliessharing to the overlapping demes of a spatially-structuredpopulation. The combination of these two methods succeeds inmaintaining multiple solutions in problems that have previouslyproved difficult for sharing alone (and vice-versa).