Start small, grow big? Saving multi-objective function evaluations

Abstract

The influence of non-constant population sizes in evolutionary multi-objective optimization algorithms is investigated. In contrast to evolutionary single-objective optimization algorithms an increasing population size is considered beneficial when approaching the Pareto-front. Firstly, different deterministic schedules are tested, featuring different parameters like the initial population size. Secondly, a simple adaptation method is proposed. Considering all results, an increasing population size during an evolutionary multi-objective optimization algorithm run saves fitness function evaluations compared to a fixed population size. In particular, the results obtained with the adaptive method are most promising.