GenSAT: A Navigational Approach

GenSAT is a family of local hill-climbing procedures for solving propositional satisfiability problems. We restate it as a navigational search process performed on an N-dimensional cube by a fictitious agent with limited lookahead. Several members of the GenSAT family have been introduced whose efficiency varies from the best in average for randomly generated problems to a complete failure on the realistic, specially constrained problems, hence raising the interesting question of understanding the essence of their different performance. In this paper, we show how we use our navigational approach to investigate this issue. We introduce new algorithms that sharply focus on specific combinations of properties of efficient GenSAT variants, and which help to identify the relevance of the algorithm features to the efficiency of local search. In particular, we argue for the reasons of higher effectiveness of HSAT compared to the original GSAT. We also derive fast approximating procedures based on variable weights that can provide good switching points for a mixed search policy. Our conclusions are validated by empirical evidence obtained from the application of several GenSAT variants to random 3SAT problem instances and to simple navigational problems.

[1]  Jeremy Frank,et al.  Weighting for Godot: Learning Heuristics for GSAT , 1996, AAAI/IAAI, Vol. 1.

[2]  Hector J. Levesque,et al.  Hard and Easy Distributions of SAT Problems , 1992, AAAI.

[3]  Toru Ishida,et al.  Improving the Learning Efficiencies of Realtime Search , 1996, AAAI/IAAI, Vol. 1.

[4]  Richard E. Korf,et al.  Real-Time Heuristic Search , 1990, Artif. Intell..

[5]  Sven Koenig Agent-Centered Search: Situated Search with Small Look-Ahead , 1996, AAAI/IAAI, Vol. 2.

[6]  Ian P. Gent,et al.  Unsatisfied Variables in Local Search , 1995 .

[7]  Hector J. Levesque,et al.  A New Method for Solving Hard Satisfiability Problems , 1992, AAAI.

[8]  Jun Gu,et al.  Efficient local search for very large-scale satisfiability problems , 1992, SGAR.

[9]  Manuela M. Veloso,et al.  Efficient Goal-Directed Exploration , 1996, AAAI/IAAI, Vol. 1.

[10]  James M. Crawford,et al.  Experimental Results on the Crossover Point inSatis ability , 1993 .

[11]  T. Ishida,et al.  Improving the Learning E ciencies of Realtime Search , 1996 .

[12]  Toby Walsh,et al.  Towards an Understanding of Hill-Climbing Procedures for SAT , 1993, AAAI.

[13]  Bart Selman,et al.  Noise Strategies for Improving Local Search , 1994, AAAI.

[14]  James M. Crawford,et al.  Experimental Results on the Crossover Point in Random 3-SAT , 1996, Artif. Intell..

[15]  Sven Koenig,et al.  Graph learning with a nearest neighbor approach , 1996, COLT '96.

[16]  Toby Walsh,et al.  An Empirical Analysis of Search in GSAT , 1993, J. Artif. Intell. Res..