Implicit Multi-Objective Coevolutionary Algorithm

Abstract

Coevolutionary algorithms are powerful tools for solving increasingly complex problems by explicitly evolving solutions in the form of interacting co-adapted subcomponents. Its fitness is subjective – fitness of individuals relative to other individuals in the population(s) – which theoretically aids the convergence rate but unfortunately births the red-queen effect. Consequently, although elitism is easy to define within a generation, it becomes hard to define across generations. Furthermore, this effect can also cause forgetfulness to occur.

We noticed that, while coevolutionary systems typically have only a single objective for evaluation, there is a subtle multi-objective aspect to evaluation that we feel necessitates a method to regulate the pairings of individuals both within and between generations. This research investigates this implicit multi-objective nature of coevolutionary systems, which, as it turns out, makes it possible to manage elitism by addressing forgetfulness and managing the Red-Queen effect, thus providing robust solutions.