-2

I'm interested in applying machine learning algorithms for SAT solving procedures. Current trends on SAT solvers seems they use CDCL procedures. Specifically, is there any small example to illustrate the idea?

hddmss
  • 231
  • 1
  • 12

1 Answers1

1

The closest work I'm aware of is by Selsam et al., "Learning a SAT Solver from Single-Bit Supervision." See https://arxiv.org/abs/1802.03685

You should also look into Selsam's PhD thesis on this topic as well https://searchworks.stanford.edu/view/13250178, which has lots of other references.

alias
  • 28,120
  • 2
  • 23
  • 40
  • Thanks for the pointers. It seems that the major SAT solvers still aren't beginning to use ML methods yet. – hddmss Aug 23 '21 at 03:54
  • 3
    SAT solving is an extremely well-established field. Unless ML has something "significant" to offer, it'll be hard to beat extremely well-tuned engines that are in existence today. I can imagine applications that arise in non-traditional fields (i.e., outside of SW/HW verification) that can give rise to problems that can benefit from these techniques, but of course this is speculation. – alias Aug 23 '21 at 05:45