Boosting combinatorial search through randomization
AAAI '98/IAAI '98 Proceedings of the fifteenth national/tenth conference on Artificial intelligence/Innovative applications of artificial intelligence
The effect of restarts on the efficiency of clause learning
IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence
Building a Hybrid SAT Solver via Conflict-Driven, Look-Ahead and XOR Reasoning Techniques
SAT '09 Proceedings of the 12th International Conference on Theory and Applications of Satisfiability Testing
Restart Strategy Selection Using Machine Learning Techniques
SAT '09 Proceedings of the 12th International Conference on Theory and Applications of Satisfiability Testing
Width-Based Restart Policies for Clause-Learning Satisfiability Solvers
SAT '09 Proceedings of the 12th International Conference on Theory and Applications of Satisfiability Testing
The Conflict-Driven Answer Set Solver clasp: Progress Report
LPNMR '09 Proceedings of the 10th International Conference on Logic Programming and Nonmonotonic Reasoning
Potassco: The Potsdam Answer Set Solving Collection
AI Communications - Answer Set Programming
Optimization methods for the partner units problem
CPAIOR'11 Proceedings of the 8th international conference on Integration of AI and OR techniques in constraint programming for combinatorial optimization problems
SAT'10 Proceedings of the 13th international conference on Theory and Applications of Satisfiability Testing
Conflict-driven answer set solving: From theory to practice
Artificial Intelligence
Factoring out assumptions to speed up MUS extraction
SAT'13 Proceedings of the 16th international conference on Theory and Applications of Satisfiability Testing
A survey of the satisfiability-problems solving algorithms
International Journal of Advanced Intelligence Paradigms
| Hi-index | 0.00 |
Most or even all competitive DPLL-based SAT solvers have a "restart" policy, by which the solver is forced to backtrack to decision level 0 according to some criterion. Although not a sophisticated technique, there is mounting evidence that this technique has crucial impact on performance. The common explanation is that restarts help the solver avoid spending too much time in branches in which there is neither an easy-to-find satisfying assignment nor opportunities for fast learning of strong clauses. All existing techniques rely on a global criterion such as the number of conflicts learned as of the previous restart, and differ in the method of calculating the threshold after which the solver is forced to restart. This approach disregards, in some sense, the original motivation of focusing on 'bad' branches. It is possible that a restart is activated right after going into a good branch, or that it spends all of its time in a single bad branch. We suggest instead to localize restarts, i.e., apply restarts according to measures local to each branch. This adds a dimension to the restart policy, namely the decision level in which the solver is currently in. Our experiments with both Minisat and Eureka show that with certain parameters this improves the run time by 15% - 30% on average (when applied to the 100 test benchmarks of SAT-race'06), and reduces the number of time-outs.