Boosting Systematic Search by Weighting Constraints

In this paper, we present a dynamic and adaptive variable ordering heuristic which guides systematic search toward inconsistent or hard parts of a Constraint Satisfaction Problem (CSP). This generic heuristic is able to exploit information about previous states of the search process whereas traditional dynamic ones only exploit information about the current state. Intuitively, it avoids some trashing by first instantiating variables involved in the constraints that have frequently participated in dead-end situations. Such information is recorded by associating a weight with each constraint. This weight is increased whenever the associated constraint is violated during search. The extensive experiments that we have conducted prove that our conflict-directed approach is the most efficient current one with respect to significant and large classes of academic, random and real-world instances.

• Type: Conference
• Year: 2004
• Where: ECAI
• Authors: Frédéric Boussemart, Fred Hemery, Christophe Lecoutre, Lakhdar Sais
• Keywords: Adaptive Variable Ordering, Artificial Intelligence, Constraint Satisfaction Problem, ECAI 2004, Traditional Dynamic Ones