Analytic study on the behavior of GAs for generalized NK-landscape problems

摘要

It is important to investigate the influence of epistasis (correlation between genes) on the behavior of Genetic Algorithms (GAs) in order to understand GAs more deeply and find ways to improve their search performance. In this work, we focus on the NK-Landscape problems, which allow us to control the amount of epistasis and its pattern, and generalize the problem in order to be closer to real-world problems. We analytically observe the behavior of GAs for the generalized problems. Also, we show the significance of an improved GA (GA-SRM) in this class of problems by a relative comparison with a canonical GA.