A Kullback-Leibler Divergence Exploration into a Look-Ahead Simulation Optimization of the Extended Compact Genetic Algorithm

Abstract

The Kullback-Leibler Divergence of gene distributions between successive generations of the Extended Compact Genetic Algorithm (ECGA) is explored. Therein, the fragility of the algorithm’s dependability to the beginning generations’ biasing is suggested. A novel approach within the scope of the ECGA for choosing a better bias by allowing the ECGA to simulate itself is presented. It is shown that, by simulating itself, the ECGA is able to use a smaller population and evaluate fewer fitness calls while maintaining the same ability to find optimal solutions.