Reproduce “Differentiable Learning of Logical Rules for Knowledge Base Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Fan Yang, Zhilin Yang, William W. Cohen
Researchers at Carnegie Mellon University developed Neural Logic Programming (Neural LP), a system for end-to-end differentiable learning of both the structure and parameters of first-order logical rules for knowledge base reasoning. This approach achieves interpretability and strong inductive generalization, demonstrating state-of-the-art results on WN18 and competitive performance across various statistical relational learning and knowledge base completion benchmarks.
Key insights
- The discrete problem of learning logical rule structures can be transformed into a continuous, differentiable optimization task through a recurrent formulation and soft attention mechanisms.
- Integrating a neural controller with a differentiable logical inference engine (TensorLog) allows for the efficient induction of interpretable, entity-independent first-order logical rules.
- Explicit logical rules recovered from the model provide interpretability, while their entity-independent nature enables robust inductive generalization to unseen entities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.