An Incremental Approach for Niching and Building Block Detection via Clustering

Abstract

Diversity preservation has already been established as an important concern for evolutionary computation. Clustering techniques were, among others, successfully applied to this purpose. Another important aspect of the research on evolutionary computation is related to linkage learning - the detection of the problem structure avoiding disruption of building blocks when new individuals are generated. This paper presents a novel approach which is a new estimation of distribution algorithm (EDA) where clustering plays two roles: diversity preservation and linkage learning. Initial empirical investigations illustrate the behavior of the algorithm when solving two benchmark optimization problems.