Longer-Term Memory in Clause Weighting Local Search for SAT

Abstract. This paper presents a comparative study between a state-ofthe-art clause weighting local search method for satisfiability testing and a variant modified to obtain longer-term memory from a global measure of clause perturbation. We present empirical evidence indicating that by learning which clauses are hardest to satisfy, the modified method can offer significant performance improvements for a well-known range of satisfiable problems. We conclude that our method’s ability to learn, and consequently to offer performance improvement, can be attributed to its ability to obtain information from a global measure of hardness, rather than from the contextual perspective exploited in previous works.

• Type: Conference
• Year: 2004
• Where: AUSAI
• Authors: Valnir Ferreira Jr., John Thornton
• Keywords: Artificial Intelligence, AUSAI 2004, Global Measure, Local Search Method, Performance Improvement