[3E1-GS-2] Machine learning: Explainable AI (1)

[3E1-GS-2-02] Learning Explainable Logical Rules through Graph Embedding

〇Yin Jun Phua^1,2, Katsumi Inoue^1,2 (1. SOKENDAI (The Graduate University for Advanced Studies), 2. National Institute of Informatics)

Keywords:interpretability, graph embedding, explainability, clustering, logic programming

Recent years have seen the surge in machine learning applications within various fields. As practitioners seeks to utilize machine learning methods in areas that affect our day-to-day lives, accountability and verification is still seen as the largest obstacle to mass adoption. Despite research advancements in the interpretability of deep learning models, the massive amount of rules generated by these methods do not allow a human to understand the models any better. To allow better understanding of huge and complex logic programs, we propose a method that utilizes graph embedding to cluster the atoms and simplifies the resulting program. We perform several experiments to prove the effectiveness of our method, and also show that the resulting program is much easier to read and understand than the original program.