Home » HACKER-TECH » NeuroSAT: Studying a SAT Solver from Single-Bit Supervision

NeuroSAT: Studying a SAT Solver from Single-Bit Supervision

(Submitted on Eleven Feb 2018)

Abstract: We most up-to-date NeuroSAT, a message passing neural network that learns to resolve
SAT concerns after only being skilled as a classifier to predict
satisfiability. Though it’s now not competitive with explain-of-the-artwork SAT
solvers, NeuroSAT can resolve concerns which shall be substantially bigger and more
complicated than it ever noticed during coaching by simply running for more
iterations. Furthermore, NeuroSAT generalizes to new distributions; after
coaching only on random SAT concerns, at test time it will perchance presumably must always resolve SAT concerns
encoding graph coloring, clique detection, dominating region, and vertex quilt
concerns, all on lots of distributions over limited random graphs.

Topics: Synthetic Intelligence (cs.AI); Studying (cs.LG); Good judgment in Pc Science (cs.LO)
Cite as: arXiv:1802.03685 [cs.AI]
  (or arXiv:1802.03685v1 [cs.AI] for this version)

Submission ancient past

From: Daniel Selsam [
bump into email]

[v1] Solar, Eleven Feb 2018 03:04:28 GMT (474kb,D)

Read Extra

Leave a Reply

Your email address will not be published. Required fields are marked *

*