GRASP: A Search Algorithm for Propositional Satisfiability
IEEE Transactions on Computers
A machine program for theorem-proving
Communications of the ACM
Chaff: engineering an efficient SAT solver
Proceedings of the 38th annual Design Automation Conference
Satometer:: how much have we searched?
Proceedings of the 39th annual Design Automation Conference
Efficient conflict driven learning in a boolean satisfiability solver
Proceedings of the 2001 IEEE/ACM international conference on Computer-aided design
SATO: An Efficient Propositional Prover
CADE-14 Proceedings of the 14th International Conference on Automated Deduction
BerkMin: A Fast and Robust Sat-Solver
Proceedings of the conference on Design, automation and test in Europe
Using CSP look-back techniques to solve real-world SAT instances
AAAI'97/IAAI'97 Proceedings of the fourteenth national conference on artificial intelligence and ninth conference on Innovative applications of artificial intelligence
| Hi-index | 0.00 |
New heuristics and strategies have enabled major advancements in SAT solving in recent years. However, experimentation has shown that there is no winning solution that works in all cases. A degradation of orders of magnitude can be observed if the wrong heuristic is chosen. The problem is that it is impossible to know, in advance, which heuristics are best for a given problem. Consequently, many ideas - those that turn out to be useful for a small subset of the cases, but significantly increase run times on most others - are discarded. We propose the notion of Adaptive Solving as a possible solution to this problem. In our framework, the SAT solver monitors the effectiveness of the search on-the-fly using a Performance Metric. The metric gives a score according to its assessment of the search progress. Based on this score, one or more heuristics are turned on or off. The goal is to use a specific heuristic or strategy when it is advantageous, and turn it off when it is not, before it does too much damage. We suggest several possible metrics, and compare their effectiveness. Our adaptive solver achieves significant speedups on a large set of examples. We also show that applying different heuristics on different parts of the search space can improve run times even beyond what can be achieved by the best heuristic on its own.