In theoretical research and practical applications, Boolean Satisfiability (SAT) is the most studied combinatorial optimization problem. Despite the fact that it is a canonical NP-complete problem which is not believed to be solvable faster than by brute-force search, in the last few decades enormous progress has been made on practical SAT solvers that are widely used for software and hardware verification, challenging algebraic problems and many more. On the other hand, machine learning and deep neural networks are opening new doors to solving data validation problems, risk management, pattern recognition, and almost every problem in the business sector day by day. A natural question is how to harness the power of neural networks to help us solve SAT, and for which classes of instances would this approach give better results than the current heuristics. This study will explore a Graph Neural Network (GNN) based SAT solver named "NeuroSAT." We looked at a variety of SAT problems to see which of them NeuroSAT can solve, and which instances remain difficult. Overall, we found that NeuroSAT can only handle very small and easily satisfiable instances, and, surprisingly, increasing the number of iterations of its message-passing stage beyond 100 does not necessarily help.
-
Notifications
You must be signed in to change notification settings - Fork 0
souravskr/MS-Project
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Experiments on Neural Network-Based SAT Solver with different instances.
Topics
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published