Divide and Conquer in Coevolution: A Difficult Balancing Act

Abstract

In recent years, Cooperative Coevolutionary Evolutionary Algorithms (CCEAs) have been developed as extensions to traditional Evolutionary Algorithms (EAs). CCEAs attempt to solve the optimization problems by decomposing them into subcomponents referred to as collaborators. CCEAs have been deemed attractive for certain complex problems (with high number of decision variables), as they can achieve better fitness values than traditional EAs by employing divide and conquer strategy. However, their performance can vary from good to bad depending on the choice of collaborators, separability of problem and the underlying recombination scheme. This chapter highlights that a basic CCEA is inadequate to handle a wide variety of problems. Thereafter, a CCEA with adaptive partitioning (CCEA-AVP) has been introduced, which attempts to chose the collaborators adaptively during the search, depending on the relationships between the design variables. Studies have been done on various test functions and the proposed technique has been compared with conventional EA as well as conventional CCEA to highlight the benefits. A number of areas of further research in CCEA are highlighted to fully exploit the benefits of coevolution.