A Coevolutionary Algorithm with Dimension Extraction

How to provide accurate evaluation in coevolution is a long-standing question. Recent work has shown that within coevolution problems, there is an implicit dimension structure in which the dimension information of an individual can uniquely determine its performance. In this paper, we propose a strict dimension extracting method. We embed it in a coevolutionary algorithm, which can automatically extract the dimensions of the problem during execution and use them to guide evaluation and selection. The experiment results show the viability of our algorithm.