Interactive (1+1) evolutionary strategy with one-fifth success rule

Abstract

Incorporation of fitness evaluation by a human user into evolutionary computation is called interactive evolutionary computation (IEC). Various IEC methods have been studied for design problems. An important challenge in IEC is to decrease human user's workload for fitness evaluation. For example, it is almost impossible for human users to continue to examine and evaluate tens of thousands of solutions. In some application fields such as evolutionary music, it is impossible to evaluate multiple solutions simultaneously. In our previous study, we formulated an IEC model by assuming the minimum level of the human user's ability to evaluate the fitness of each solution. We also illustrated our IEC model through computational experiments on combinatorial optimization problems. In this study, we address the use of our IEC model for continuous optimization problems. We propose an idea to incorporate step-size adaptation by the well-known one-fifth success rule into our IEC model. Through computational experiments on four test problems, we examine the search ability of our IEC model with the step-size adaptation mechanism.