Conflict Detection for Arbitrary Constraint Propagation Algorithms

Existing conflict detection methods for CSP’s such as [de Kleer, 1989; Ginsberg, 1993 ] cannot make use of powerful propagation which makes them unusable for complex real-world problems. On the other hand, powerful constraint propagation methods lack the ability to extract dependencies or conflicts, which makes them unusable for many advanced AI reasoning methods that require conflicts, as well as for interactive applications that require explanations. In this paper, we present a non-intrusive conflict detection algorithm called QUICKXPLAIN that tackles those problems. It can be applied to any propagation or inference algorithm as powerful as it may be. Our algorithm improves the efficiency of direct non-intrusive conflict detectors by recursively partitioning the problem into subproblems of half the size and by immediately skipping those subproblems that do not contain an element of the conflict. Q UICKXPLAIN is used as explanation component of an advanced industrial constraint-based configuration tool.