Injecting CMA-ES into MOEA/D

Abstract

MOEA/D is an aggregation-based evolutionary algorithm whichhas been proved extremely efficient and effective for solving multi-objective optimization problems. It is based on the idea of de-composing the original multi-objective problem into several single-objective subproblems by means of wel l-defined scalari zi ng f unc-tions. Those single-objective subproblems are solved in a cooper-ative manner by defining a neighborhood relation between them.This makes MOEA/D particularly interesting when attempting toplug and to leverage single-objective optimizers in a multi-objectivesetting. In this context, we investigate the benefits that MOEA/Dcan achieve when coupled with CMA-ES, which is believed to bea pow erful single-objective optimizer. We rely on the ability ofCMA-ES to deal with injected solutions in order to update differ-ent covariance matrices with respect to each subproblem definedin MOEA/D. We show that by cooperatively evolving neighboringCMA-ES components, we are able to obtain competitive results fordifferent multi-objective benchmark functions.