A model of anytime algorithm performance for bi-objective optimization - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Article Dans Une Revue Journal of Global Optimization Année : 2021

A model of anytime algorithm performance for bi-objective optimization

Résumé

Anytime algorithms allow a practitioner to trade-off runtime for solution quality. This is of particular interest in multi-objective combinatorial optimization since it can be infeasible to identify all efficient solutions in a reasonable amount of time. We present a theoretical model that, under some mild assumptions, characterizes the “optimal” trade-off between runtime and solution quality, measured in terms of relative hypervolume, of anytime algorithms for bi-objective optimization. In particular, we assume that efficient solutions are collected sequentially such that the collected solution at each iteration maximizes the hypervolume indicator, and that the non-dominated set can be well approximated by a quadrant of a superellipse. We validate our model against an “optimal” model that has complete knowledge of the non-dominated set. The empirical results suggest that our theoretical model approximates the behavior of this optimal model quite well. We also analyze the anytime behavior of an ε-constraint algorithm, and show that our model can be used to guide the algorithm and improve its anytime behavior.
Fichier principal
Vignette du fichier
main.pdf (1.95 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-02898963 , version 1 (27-07-2020)

Identifiants

Citer

Alexandre Borges de Jesus, Luis Paquete, Arnaud Liefooghe. A model of anytime algorithm performance for bi-objective optimization. Journal of Global Optimization, 2021, 79, pp.329-350. ⟨10.1007/s10898-020-00909-9⟩. ⟨hal-02898963⟩
143 Consultations
169 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More