Explainability Techniques for Graph Convolutional Networks

Abstract

Graph Networks are used to make decisions in potentially complex scenarios but it is usually not obvious how or why they made them. In this work, we study the explainability of Graph Network decisions using two main classes of techniques, gradient-based and decomposition-based, on a toy dataset and a chemistry task. Our study sets the ground for future development as well as application to real-world problems.

Accepted at the ICML 2019 Workshop "Learning and Reasoning with Graph-Structured Representations" (poster + spotlight talk)

References in corpus (4)

• Variational Graph Auto-Encoders
• SmoothGrad: removing noise by adding noise
• Interaction Networks for Learning about Objects, Relations and Physics
• A Compositional Object-Based Approach to Learning Physical Dynamics