Learning CNF formulas from uniform random solutions in the local lemma regime
By: Weiming Feng , Xiongxin Yang , Yixiao Yu and more
Potential Business Impact:
Teaches computers to understand complex logic puzzles.
We study the problem of learning a $n$-variables $k$-CNF formula $\Phi$ from its i.i.d. uniform random solutions, which is equivalent to learning a Boolean Markov random field (MRF) with $k$-wise hard constraints. Revisiting Valiant's algorithm (Commun. ACM'84), we show that it can exactly learn (1) $k$-CNFs with bounded clause intersection size under Lov\'asz local lemma type conditions, from $O(\log n)$ samples; and (2) random $k$-CNFs near the satisfiability threshold, from $\widetilde{O}(n^{\exp(-\sqrt{k})})$ samples. These results significantly improve the previous $O(n^k)$ sample complexity. We further establish new information-theoretic lower bounds on sample complexity for both exact and approximate learning from i.i.d. uniform random solutions.
Similar Papers
Learning Juntas under Markov Random Fields
Machine Learning (CS)
Teaches computers to learn patterns from data.
Searching for Falsified Clause in Random (log n)-CNFs is Hard for Randomized Communication
Computational Complexity
Finds why a puzzle has no answer.
Approximate Counting in Local Lemma Regimes
Data Structures and Algorithms
Counts hard-to-count things much faster.