Learn to Explain Efficiently via Neural Logic Inductive Learning
It addresses the need for interpretable and self-explanatory decisions in responsible machine learning systems, representing an incremental improvement in efficiency and scalability for logic-based methods.
The paper tackles the problem of learning interpretable first-order logic rules efficiently in inductive logic programming, achieving rules that are 10 times longer and 3 times faster than state-of-the-art methods, and scaling to large datasets like Visual Genome with 1 million entities.
The capability of making interpretable and self-explanatory decisions is essential for developing responsible machine learning systems. In this work, we study the learning to explain problem in the scope of inductive logic programming (ILP). We propose Neural Logic Inductive Learning (NLIL), an efficient differentiable ILP framework that learns first-order logic rules that can explain the patterns in the data. In experiments, compared with the state-of-the-art methods, we find NLIL can search for rules that are x10 times longer while remaining x3 times faster. We also show that NLIL can scale to large image datasets, i.e. Visual Genome, with 1M entities.