Multiwinner voting in genetic algorithms for solving ill-posed global optimization problems

Abstract

Genetic algorithms are a group of powerful tools for solving ill-posed global optimization problems in continuous domains. In case in which the insensitivity of the fitness function is the main obstacle, the most desired feature of a genetic algorithm is its ability to explore plateaus of the fitness function, surrounding its minimizers. In this paper we suggest a way of maintaining diversity of the population in the plateau regions, based on a new approach for the selection based on the theory of multiwinner elections among autonomous agents. The paper delivers a detailed description of the new selection algorithm, computational experiments that guide the choice of the proper multiwinner rule to use, and a preliminary experiment showing the proposed algorithm’s effectiveness in exploring a fitness function’s plateau.