ANDRE: Attention-based Neuro-symbolic Differentiable Rule Extractor
A novel framework named ANDRE (Attention-based Neuro-symbolic Differentiable Rule Extractor) has been introduced for Inductive Logic Programming (ILP). This system acquires first-order logic programs by optimizing within a continuous rule space, utilizing attention-based logical operators. ANDRE substitutes traditional rule templates and logical operators with fully differentiable conjunction and disjunction operators that mimic logical min-max semantics. This innovation tackles scalability challenges in noisy and probabilistic environments, where the conventional discrete combinatorial search of ILP is fragile, and differentiable techniques often encounter issues like vanishing gradients or inadequate approximations. The methodology is elaborated in a paper available on arXiv (2605.04193).
Key facts
- ANDRE stands for Attention-based Neuro-symbolic Differentiable Rule Extractor.
- It is a framework for Inductive Logic Programming (ILP).
- ANDRE learns first-order logic programs by optimizing over a continuous rule space.
- It uses attention-based logical operators for conjunction and disjunction.
- The operators approximate logical min-max semantics.
- Classical ILP relies on discrete combinatorial rule search and is brittle under uncertainty.
- Differentiable ILP methods typically depend on predefined rule templates or inaccurate fuzzy operators.
- The paper is available on arXiv with identifier 2605.04193.
Entities
Institutions
- arXiv