A closer look to elitism in ε-dominance many-objective optimization

Abstract

Elitism is a common feature of many-objective optimizers and has a strong impact on the performance of the algorithms. The way elitism is implemented vary among the various approaches to many-objective optimization and there are no detailed studies about their effects. In this work we focus on a multi- and many-objective optimization approach based on ε-dominance. We track the number of generations a solution remains in the population to bias survival selection or the creation of neighborhoods for parent selection. We investigate how elitist strategies affect performance of the algorithm and show that convergence and diversity can be enhanced by using different strategies for elitism on many-objective uni-modal and multi-modal problems with 4, 5, and 6 objectives.