Exponential Propagation for Set Variables

Research on constraint propagation has primarily focused on designing polynomial-time propagators sometimes at the cost of a weaker filtering. Interestingly, the evolution of constraint programming over sets have been diametrically different. The domain representations are becoming increasingly expensive computationally and theoretical results appear to question the wisdom of these research directions. This paper explores this apparent contradiction by pursuing even more complexity in the domain representation and the filtering algorithms. It shows that the product of the length-lex and subset-bound domains improves filtering and produces orders of magnitude improvements over existing approaches on standard benchmarks. Moreover, the paper proposes exponential-time algorithms for NP-hard intersection constraints and demonstrates that they bring significant performance improvements and speeds up constraint propagation considerably.

• Type: Journal
• Year: 2010
• Where: CP
• Authors: Justin Yip, Pascal Van Hentenryck
• Keywords: Artificial Intelligence, Constraint Propagation, CP 2010, Domain Representation, Polynomial-time Propagators