Inverse Consistencies for Non-Binary Constraints

We present a detailed study of two inverse consistencies for non-binary constraints: relational path inverse consistency (rel PIC) and pairwise inverse consistency (PWIC). These are stronger than generalized arc consistency (GAC), even though they also only prune domain values. We propose algorithms to achieve rel PIC and PWIC, that have a time complexity better than the previous generic algorithm for inverse consistencies. One of our algorithms for PWIC has a complexity comparable to that for GAC despite doing more pruning. Our experiments demonstrate that inverse consistencies can be more efficient than GAC on a range of non-binary problems.

• Type: Conference
• Year: 2006
• Where: ECAI
• Authors: Kostas Stergiou, Toby Walsh
• Keywords: Artificial Intelligence, ECAI 2006, Inverse Consistencies, Inverse Consistency, Rel Pic