MatthiasNickles/diff-SAT

Embeds gradient descent directly into CDNL-based SAT/ASP solving to optimize differentiable cost functions over model multisets, enabling probabilistic inference without independence assumptions. Supports weighted Boolean variables, clauses, and rules to integrate probabilistic and logical constraints, with output as probability distributions over possible worlds. Runs on the Java VM with multithreading and handles both standard SAT/ASP solving and probabilistic satisfiability (PSAT/CPA) problems with distribution-aware model sampling.

No commits in the last 6 months.