Proceedings Abstracts of the Twenty-Fourth International Joint Conference on Artificial Intelligence

MergeXplain: Fast Computation of Multiple Conflicts for Diagnosis / 3221
Kostyantyn Shchekotykhin, Dietmar Jannach, Thomas Schmitz

The computation of minimal conflict sets is a central task when the goal is to find relaxations or explanations for overconstrained problem formulations and in particular in the context of Model-Based Diagnosis (MBD) approaches. In this paper we propose MergeXPlain, a non-intrusive conflict detection algorithm which implements a divide-and-conquer strategy to decompose a problem into a set of smaller independent subproblems. Our technique allows us to efficiently determine multiple minimal conflicts during one single problem decomposition run, which is particularly helpful in MBD problem settings. An empirical evaluation on various benchmark problems shows that our method can lead to a significant reduction of the required diagnosis times.