Controlling the cooperative-competitive boundary in niched genetic algorithms

Niching can allow a diverse population to cooperatively represent a single, distributed solution to the problem at hand. Success- ful niching mechanisms must promote both cooperation (i.e., co;existence of separate "species" for each desired niche), and competition (i.e., intensive search for the best species for each niche). In this paper we seek to control the competitive-cooperative boundary in the space of possible niche relationships, so that we can choose which pairs of interacting niches will survive under GA selection and which niche pairs will be resolved to yield a single winner. We introduce the concept of resource replenishment period, tau, 85 a control on the relative importance of objective fitness over-diversity pressure. We find that by varying tau between zero and one, we can smoothly transition between pure selection and full niching.