Adaptive Selection of Helper-Objectives with Reinforcement Learning

Abstract

In this paper a previously proposed method of choosing auxiliary fitness functions is applied to adaptive selection of helper-objectives. Helper-objectives are used in evolutionary computation to enhance the optimization of the primary objective. The method based on choosing between objectives of a single-objective evolutionary algorithm with reinforcement learning is briefly described. It is tested on a model problem. From the results of the experiment, it can be concluded that the method allows to automatically select the most effective helper-objectives and ignore the ineffective ones. It is also shown that the proposed method outperforms multi-objective evolutionary algorithms, that were used with helper-objectives originally.