Reducing the Cost of Exploring Neighborhood Areas in Dynamic Local Search for SAT
Abdelraouf Ishtaiwi *
Faculty of IT, School of CS, University of Petra, Amman, Jordan.
Ghassan Issa
Faculty of IT, School of CS, University of Petra, Amman, Jordan.
Wael Hadi
Faculty of IT, School of CS, University of Petra, Amman, Jordan.
*Author to whom correspondence should be addressed.
Abstract
Stochastic Local Search (SLS) algorithms are of great importance to many fields of Computer Sciences and Artificial Intelligence. This is due to their efficient performance when applied for solving randomly generated satisfiability problems (SAT). Our focus in the current work is on one of the SLS dynamic weighting approaches known as multi-level weight distribution (mulLWD). We experimentally investigated the performance and the weight behaviors of mulLWD. Based on our experiments, we observed that the 2nd level weights movements could lead to poor performance of mulLWD, especially when applied for solving large and harder SAT problems. Therefore, we developed a new heuristic that could reduce the cost of the 2nd level neighborhood exploitation known as partial multi-level weight distribution mulLWD+. Experimental results indicate that mulLWD+ heuristic has significantly better performance than mulLWD in a wide range of SAT problems.
Keywords: Artificial intelligence, Boolean satisfiability, search algorithms