Social Attraction Mutation: A Novel Method for Mutation Based on Attraction

Márk Domonkos, Zhang Huanpeng, Natabara Máté Gyöngyössy, János Botzheim

Abstract: In the field of Computational Intelligence, Evolutionary Algorithms are considered a widely known and used group of methods. In this paper, we present a novel type of mutation operator, called Social Attraction Mutation, based on Random Deviation Mutation and inspired by the Social Attraction of belonging to a leader group. We carry out tests on our operator with a wide range of parameter sets and execution configurations, on four continuous benchmark functions, and two classical control problems from Gymnasium (successor of OpenAI Gym). The results indicate that this kind of mutation operation can lead to better convergence during the search. However, it is too greedy to use it alone, without combining it with another, probabilistic mutation operator.