Parallel Constraint-Driven Inductive Logic Programming

Andrew Cropper,Oghenejokpeme Orhobor,Cristian Dinu,Rolf Morel
DOI: https://doi.org/10.48550/arXiv.2109.07132
2021-09-15
Abstract:Multi-core machines are ubiquitous. However, most inductive logic programming (ILP) approaches use only a single core, which severely limits their scalability. To address this limitation, we introduce parallel techniques based on constraint-driven ILP where the goal is to accumulate constraints to restrict the hypothesis space. Our experiments on two domains (program synthesis and inductive general game playing) show that (i) parallelisation can substantially reduce learning times, and (ii) worker communication (i.e. sharing constraints) is important for good performance.
Artificial Intelligence,Machine Learning
What problem does this paper attempt to address?