Dead-end elimination for weighted CSP - INRAE - Institut national de recherche pour l’agriculture, l’alimentation et l’environnement Accéder directement au contenu
Communication Dans Un Congrès Année : 2013

Dead-end elimination for weighted CSP

Résumé

Soft neighborhood substitutability (SNS) is a powerful technique to automatically detect and prune dominated solutions in combinatorial optimization. Recently, it has been shown in [26] that enforcing partial SNS (PSNSr) during search can be worthwhile in the context of Weighted Constraint Satisfaction Problems (WCSP). However, for some problems, especially with large domains, PSNSr is still too costly to enforce due to its worst-case time complexity in O(ned4) for binary WCSP. We present a simplified dominance breaking constraint, called restricted dead-end elimination (DEEr), the worst-case time complexity of which is in O(ned2). Dead-end elimination was introduced in the context of computational biology as a preprocessing technique to reduce the search space [13, 14, 16, 17, 28, 30]. Our restriction involves testing only one pair of values per variable instead of all the pairs, with the possibility to prune several values at the same time. We further improve the original dead-end elimination criterion, keeping the same time and space complexity as DEEr. Our results show that maintaining DEEr during a depth-first branch and bound (DFBB) search is often faster than maintaining PSNSr and always faster than or similar to DFBB alone.
Fichier principal
Vignette du fichier
Givry13a_1.pdf (242.81 Ko) Télécharger le fichier
Origine : Fichiers éditeurs autorisés sur une archive ouverte
Loading...

Dates et versions

hal-02746292 , version 1 (03-06-2020)

Identifiants

Citer

Simon de Givry, Steve Prestwich, Barry O'Sullivan. Dead-end elimination for weighted CSP. 19th International Conference, Constraint Programming 2013, Sep 2013, Uppsala, Sweden. 10 p., ⟨10.1007/978-3-642-40627-0_22⟩. ⟨hal-02746292⟩
6 Consultations
72 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More