Titre :

Studying MOEAs Dynamics and their Performance using a Three Compartmental Model

Auteur(s) :

Monzón, Hugo [Auteur]
Faculty of Engineering [Nagano]
Aguirre, Hernan [Auteur]
Faculty of Engineering [Nagano]
Verel, Sébastien [Auteur]
Laboratoire d'Informatique Signal et Image de la Côte d'Opale [LISIC]
Liefooghe, Arnaud [Auteur]
Optimisation de grande taille et calcul large échelle [BONUS]
Derbel, Bilel [Auteur]
Optimisation de grande taille et calcul large échelle [BONUS]
Tanaka, Kiyoshi [Auteur]
Faculty of Engineering [Nagano]

Titre de la manifestation scientifique :

GECCO 2018 - Genetic and Evolutionary Computation Conference Companion

Ville :

Kyoto

Pays :

Japon

Date de début de la manifestation scientifique :

2018-07-15

Mot(s)-clé(s) en anglais :

Empirical study
Working principles of evolutionary computing
Genetic algorithms
Multi-objective optimization

Discipline(s) HAL :

Informatique [cs]/Intelligence artificielle [cs.AI]

Résumé en anglais : [en]

The road to a better design of multi- and many-objective evolutionary algorithms requires a deeper understanding of their behavior. A step on this road has recently been taken with the proposal of compartmental models to ...
Lire la suite >The road to a better design of multi- and many-objective evolutionary algorithms requires a deeper understanding of their behavior. A step on this road has recently been taken with the proposal of compartmental models to study population dynamics. In this work, we push this step further by introducing a new set of features that we link with algorithm performance. By tracking the number of newly discovered Pareto Optimal (PO) solutions, the previously-found PO solutions and the remaining non-PO solutions, we can track the algorithm progression. By relating these features with a performance measure, such as the hypervolume, we can analyze their relevance for algorithm comparison. This study considers out-of-the-box implementations of recognized multi- and many-objective optimizers belonging to popular classes such as conventional Pareto dominance, extensions of dominance, indicator, and decomposition based approaches. In order to generate training data for the compartmental models, we consider multiple instances of MNK-landscapes with different numbers of objectives.Lire moins >

Langue :

Anglais

Comité de lecture :

Oui

Audience :

Internationale

Vulgarisation :

Non

Collections :

• Centre de Recherche en Informatique, Signal et Automatique de Lille (CRIStAL) - UMR 9189

Source :

Harvested from HAL