Pareto dominance-based MOEAs on Problems with Difficult Pareto Set Topologies

Yuri Marca (1) , Hernan Aguirre (2) , Saúl Zapotecas-Martínez (2) , Arnaud Liefooghe (3) , Sébastien Verel (4) , Bilel Derbel (3) , Kiyoshi Tanaka (2)

1 Shinshu University [Nagano]
2 Faculty of Engineering [Nagano]
3 BONUS - Optimisation de grande taille et calcul large échelle
4 LISIC - Laboratoire d'Informatique Signal et Image de la Côte d'Opale

Despite the extensive application of multi-objective evolutionary algorithms (MOEAs) to solve multi-objective optimization problems (MOPs), understanding their working principles is still open to research. One of the most popular and successful MOEA approaches is based on Pareto dominance and its relaxed version, Pareto epsilon-dominance. However, such approaches have not been sufficiently studied in problems of increased complexity. In this work, we study the effects of the working mechanisms of the various components of these algorithms on test problems with difficult Pareto set topologies. We focus on separable unimodal and multimodal functions with 2, 3, and 4 objectives, all having difficult Pareto set topologies. Our experimental study provides some interesting and useful insights to understand better Pareto dominance-based MOEAs.

Mots clés: Working principles of evolutionary computing; Multi-objective optimization; Selection; Recombination operators; Differential Evo- lution.

Domaines: Intelligence artificielle [cs.AI]

Yuri Marca, Hernan Aguirre, Saúl Zapotecas-Martínez, Arnaud Liefooghe, Sébastien Verel, et al.. Pareto dominance-based MOEAs on Problems with Difficult Pareto Set Topologies. GECCO 2018 - Genetic and Evolutionary Computation Conference Companion, Jul 2018, Kyoto, Japan. ⟨hal-01823715⟩