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

A MaxSAT Algorithm Using Cardinality Constraints of Bounded Size / 2677
Mario Alviano, Carmine Dodaro, Francesco Ricca

Core-guided algorithms proved to be effective on industrial instances of MaxSAT, the optimization variant of the satisfiability problem for propositional formulas. These algorithms work by iteratively checking satisfiability of a formula that is relaxed at each step by using the information provided by unsatisfiable cores. The paper introduces a new core-guided algorithm that adds cardinality constraints for each detected core, but also limits the number of literals in each constraint in order to control the number of refutations in subsequent satisfiability checks. The performance gain of the new algorithm is assessed on the industrial instances of the 2014 MaxSAT Evaluation.