Performance Improvement with Prior-validation Framework for Algorithmic Configuration on Self-adaptive Differential Evolution

Abstract

Self-adaptive differential evolution approaches (self-adaptive DEs) often suffer to boost their performances under a limited number of fitness evaluations, since they heavily rely on the trial-and-error process required to adapt algorithmic configurations. In order to enhance the performance in early generations, this paper presents a generalized prior-validation framework for algorithmic configurations, which can be applicable to major variants of self-adaptive DEs that adapt the scaling factor, the crossover rate, and/or the mutation/crossover strategies for each individual. Experimental results on benchmark problems show that the proposed method successfully boosts the performances of jDE, SaDE, and JADE. Thus, the proposed method reveals a possibility of self-adaptive DEs toward computationally-expensive optimization problems where self-adaptive DEs have had a difficulty.

Kei Nishihara

2nd-year Doctoral Student

My research interests include evolutionary computation and swarm intelligence.