Resolution Enhanced SLS solvers: R+PAWS, R+RSAPS and R+ANOV+

Recent work on Stochastic Local Search (SLS) for the SAT and CSP domains has shown the superior performance of SLS over traditional backtracking algorithms on a broad range of problem instances. In this paper, we report on a technique for enhancing the performance of SLS solvers by incorporating a preprocessing phase in which resolution is used to deduce consequences of the input clauses, exposing hidden structure in the problems which the solvers are then able to exploit. In the next section we describe the resolution procedure and briefly review some of its uses in SAT solvers. In the subsequent section, we report the resolution enhanced procedure for SLS solvers and outline three enhanced SLS solvers, namely R+ANOV, R+PAWS, and R+RSAPS, based on Adaptive Novelty [Hoos, 2002], PAWS [Thornton et al., 2004] and RSAPS [Hutter et al., 2002].