Enhancing SAT Solving with GNN for Clause Weight Prediction

Boolean satisfiability (SAT) solving is foundational in various computational domains but faces scalability challenges with large instances. This paper introduces a novel method employing Graph Neural Networks (GNNs) to predict clause weights dynamically within SAT solvers. By representing SAT problems as bipartite graphs and integrating GNNs for clause weight prediction, our approach achieves a 40% reduction in solving time and a 35% improvement in solution quality compared to state-of-the-art solvers such as MapleSAT and GAN-SAT. Extensive experiments on benchmarks and real-world applications validate the effectiveness of our method, indicating significant potential for enhancing SAT solver performance in practical settings.