Learning logic programs by finding minimal unsatisfiable subprograms
Andrew Cropper, Céline Hocquette
TL;DR
The paper tackles the challenge of search-space explosion in inductive logic programming by introducing Minimal Unsatisfiable Subprograms (MUSPs) as pruning anchors. It presents MUSPer, an ILP system built on Learning From Failures (LFF) and the Popper framework that detects MUSPs during learning and derives constraints to prune both specialisations and redundancies, while preserving optimality. The authors prove soundness and effectiveness of MUSP-based pruning, implement MUSPer, and demonstrate across diverse domains that learning times can be reduced dramatically (up to 99%) with equal or improved predictive accuracy. This work bridges constraint satisfaction techniques with ILP, enabling scalable learning of definite and recursive programs and offering practical speedups without sacrificing solution quality.
Abstract
The goal of inductive logic programming (ILP) is to search for a logic program that generalises training examples and background knowledge. We introduce an ILP approach that identifies minimal unsatisfiable subprograms (MUSPs). We show that finding MUSPs allows us to efficiently and soundly prune the search space. Our experiments on multiple domains, including program synthesis and game playing, show that our approach can reduce learning times by 99%.