Niche Radius Adaptation with Asymmetric Sharing

In the field of Genetic Algorithms, nichingtechniques have been invented with the aim to induce speciationon multimodal fitness landscapes. Unfortunately, they often rely on a problem-dependent niche radiusparameter. This is the niche radius problem. In recent research, the possibilities to transfer niching techniques to the field of Evolution Strategies(ES) have been studied. First attempts were carried out to learn a good value for the niche radius through self-adaptation. In this paper we introduce a new niching method for ES with self-adaptation of the niche radius: asymmetric sharing. It is a form of fitness sharing. In contrast to earlier studies, it does not depend on coupling the niche radius to other strategy parameters. Experimental results indicate that asymmetric sharing performs well in comparison to traditional sharing, without relying on problem-dependent parameters.