Better Caching for Better Model Counting (extended abstract)
State-of-the-art model counters provide a variety of branching heuristics, aiding users in configuring these solvers so they perform well on specific types of problem instances. However, they provide a limited choice of cache management strategies. We argue that the state of the art in model counting could benefit from more sophisticated heuristics for cache management. We motivate this with preliminary results and propose to use machine learning to develop new cache management heuristics.
Published in International Workshop on Model Counting (MCW), held in conjunction with SAT 2020, 2020