Titre :

Towards landscape-aware automatic algorithm configuration: preliminary experiments on neutral and rugged landscapes

Auteur(s) :

Liefooghe, Arnaud [Auteur]
Parallel Cooperative Multi-criteria Optimization [DOLPHIN]
Derbel, Bilel [Auteur]
Parallel Cooperative Multi-criteria Optimization [DOLPHIN]
Verel, Sébastien [Auteur]
Laboratoire d'Informatique Signal et Image de la Côte d'Opale [LISIC]
Aguirre, Hernan [Auteur]
Faculty of Engineering [Nagano]
Tanaka, Kiyoshi [Auteur]
Faculty of Engineering [Nagano]

Éditeur(s) ou directeur(s) scientifique(s) :

Bin Hu
Manuel López-Ibáñez

Titre de la manifestation scientifique :

EvoCOP 2017 - Conference on Evolutionary Computation in Combinatorial Optimisation

Ville :

Amsterdam

Pays :

Pays-Bas

Date de début de la manifestation scientifique :

2017-04-19

Titre de l’ouvrage :

Evolutionary Computation in Combinatorial Optimization17th European Conference, EvoCOP 2017, Amsterdam, The Netherlands, April 19-21, 2017, Proceedings

Titre de la revue :

Lecture Notes in Computer Science (LNCS)

Éditeur :

Springer

Date de publication :

2017-03

Discipline(s) HAL :

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

Résumé en anglais : [en]

The proper setting of algorithm parameters is a well-known issue that gave rise to recent research investigations from the (offline) automatic algorithm configuration perspective. Besides, the characteristics of the target ...
Lire la suite >The proper setting of algorithm parameters is a well-known issue that gave rise to recent research investigations from the (offline) automatic algorithm configuration perspective. Besides, the characteristics of the target optimization problem is also a key aspect to elicit the behavior of a dedicated algorithm, and as often considered from a landscape analysis perspective. In this paper, we show that fitness landscape analysis can open a whole set of new research opportunities for increasing the effectiveness of existing automatic algorithm configuration methods. Specifically, we show that using landscape features in iterated racing both (i) at the training phase, to compute multiple elite configurations explicitly mapped with different feature values, and (ii) at the production phase, to decide which configuration to use on a feature basis, provides significantly better results compared against the standard landscape-oblivious approach. Our first experimental investigations on NK-landscapes, considered as a benchmark family having controllable features in terms of ruggedness and neutrality, and tackled using a memetic algorithm with tunable population size and variation operators, show that a landscape-aware approach is a viable alternative to handle the heterogeneity of (black-box) combinatorial optimization problems.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