• Citer

Anytime Local Search for Multi-Objective Combinatorial Optimization: Design, Analysis and Automatic Configuration

par Dubois-Lacoste, Jérémie
Président du jury Dorigo, Marco
Promoteur Stützle, Thomas
Publication Non publié, 2014-05-26

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

In this thesis, we are interested in the design and study of heuristic algorithms for multi-objective optimization. Two aspects are key to our work and make it novel in the multi-objective optimization field: the use of automatic configuration techniques, and the consideration of the anytime behavior. Automatic configuration techniques free the human designer from the time-consuming task of setting parameters, and have already been successfully applied to single-objective algorithms. As we show in this thesis, they are both applicable and desirable for multi-objective algorithms as well. The anytime behavior of an algorithm is its ability to return as high-quality solutions as possible at any moment of its execution. Having algorithms that have a good anytime behavior is highly desirable in situations where the computation time is unknown a priori or when it changes each time the algorithm is launched. The optimization literature is mostly focused on the quality reached by an algorithm after a given time defined a priori, and the outcomes of such studies may poorly extend to situations where the anytime behavior is relevant. In this thesis, besides studying the anytime behavior of multi-objective algorithms, we also test the application of automatic configuration techniques in order to obtain automatically good “anytime” algorithms. All algorithms are evaluated on well-known combinatorial problems by means of statistical and graphical tools. Our results, published in several international journals and conferences, improve over thestate-of-the-art for relevant and widely studied problems, in terms of anytime behavior and also in terms of final quality.