Search Ordering Heuristics for Restarts-Based Constraint Solving

Constraint Satisfaction Problems are ubiquitous in Artificial Intelligence. Over the past decade significant advances have been made in terms of the size of problem instance that can be solved due to insights gained from the study of runtime distributions of systematic backtrack search algorithms. A particularly impressive advance has been the use of randomization and restarts, which are now a standard component of state-of-the-art solvers. In this paper we propose a new class of variable and value ordering heuristics that learn from nogoods without storing them. The empirical analysis provides clear evidence that the proposed ordering heuristics dramatically improve the performance of restarts-based constraint solving. We can regard our heuristics as exploiting positive experience to improve search.

• Type: Conference
• Year: 2007
• Where: FLAIRS
• Authors: Margarita Razgon, Barry O'Sullivan, Gregory M. Provan
• Keywords: Artificial Intelligence, Constraint Satisfaction Problems, FLAIRS 2007, Ordering Heuristics, Value Ordering Heuristics