Evolving Fitness Landscapes with Complementary Fitness Functions

Vincent Hénaux, Adrien Goëffon, Frédéric Saubion

Abstract: Given an optimization problem, local search algorithms may fail to reach optimal solutions when faced to difficult and unsuitable fitness landscapes. Climbing based optimization is sensitive to unexpected distribution of local optima. In this paper, we aim at modifying the initial fitness landscape of a problem in order to better fit climbing requirements. We propose thus a fitness landscape generation framework based on an evolutionary process. Preliminary experiments are presented as a proof of concept.