AAAI Publications, Twenty-Fourth AAAI Conference on Artificial Intelligence

Font Size: 
A Stronger Consistency for Soft Global Constraints in Weighted Constraint Satisfaction
Jimmy Lee, K. L. Leung

Last modified: 2010-07-03

Abstract


Weighted Constraint Satisfaction is made practical by powerful consistency techniques, such as AC*, FDAC* and EDAC*, which reduce search space effectively and efficiently during search, but they are designed for only binary and ternary constraints. To allow soft global constraints, usually of high arity, to enjoy the same benefits, Lee and Leung give polynomial time algorithms to enforce generalized AC* (GAC*) and FDAC* (FDGAC*) for projection-safe soft non-binary constraints. Generalizing the stronger EDAC* is less straightforward. In this paper, we first reveal the oscillation problem when enforcing EDAC* on constraints sharing more than one variable. To avoid oscillation, we propose a weak version of EDAC* and generalize it to weak EDGAC* for non-binary constraints. Weak EDGAC* is stronger than FDGAC* and GAC*, but weaker than VAC and soft k-consistency for k > 2. We also show that weak EDGAC* can be enforced in polynomial time for projection-safe constraints. Extensive experimentation confirms the efficiency of our proposal.

Full Text: PDF