Adaptive Parameterized Consistency for Non-Binary CSPs by Counting Supports

Robert J. Woodward (1) , Anthony Schneider (1) , Berthe Y. Choueiry (1) , Christian Bessiere (2)

1 Constraint Systems Laboratory
2 COCONUT - Agents, Apprentissage, Contraintes

Determining the appropriate level of local consistency to en- force on a given instance of a Constraint Satisfaction Problem (CSP) is not an easy task. However, selecting the right level may determine our ability to solve the problem. Adaptive parameterized consistency was recently proposed for binary CSPs as a strategy to dynamically select one of two local consistencies (i.e., AC and maxRPC). In this paper, we propose a similar strategy for non-binary table constraints to select between enforcing GAC and pairwise consistency. While the former strategy approximates the supports by their rank and requires that the variablesndomains be ordered, our technique removes those limitations. We empirically evaluate our approach on benchmark problems to establish its advantages.

Domaines: Intelligence artificielle [cs.AI]

Robert J. Woodward, Anthony Schneider, Berthe Y. Choueiry, Christian Bessiere. Adaptive Parameterized Consistency for Non-Binary CSPs by Counting Supports. CP: Principles and Practice of Constraint Programming, Sep 2014, Lyon, France. pp.755-764, ⟨10.1007/978-3-319-10428-7_54⟩. ⟨lirmm-01067342⟩