Differentiable Rule Induction from Raw Sequence Inputs
Kun Gao, Katsumi Inoue, Yongzhi Cao, Hanpin Wang, Feng Yang
TL;DR
This work tackles the problem of extracting interpretable logical rules directly from raw data without explicit label leakage. It introduces NeurRL, a fully differentiable neuro-symbolic ILP framework that combines autoencoder-based representations, differentiable clustering, and a neural rule-learning module to induce rules from raw sequences and images. By formalizing ILP from raw inputs via interpretation transitions and using a program tensor $\mathbf{M}_P$ to encode rules, NeurRL achieves competitive accuracy on time-series benchmarks and provides interpretable rule bodies built from pattern-region predicates. The results demonstrate effective rule discovery with precise and recallful explanations, along with favorable training efficiency and robust ablations, suggesting practical impact for explainable AI in domains like healthcare and surveillance.
Abstract
Rule learning-based models are widely used in highly interpretable scenarios due to their transparent structures. Inductive logic programming (ILP), a form of machine learning, induces rules from facts while maintaining interpretability. Differentiable ILP models enhance this process by leveraging neural networks to improve robustness and scalability. However, most differentiable ILP methods rely on symbolic datasets, facing challenges when learning directly from raw data. Specifically, they struggle with explicit label leakage: The inability to map continuous inputs to symbolic variables without explicit supervision of input feature labels. In this work, we address this issue by integrating a self-supervised differentiable clustering model with a novel differentiable ILP model, enabling rule learning from raw data without explicit label leakage. The learned rules effectively describe raw data through its features. We demonstrate that our method intuitively and precisely learns generalized rules from time series and image data.