Explainability Techniques for Graph Convolutional Networks

TL;DR

This paper reviews explainability techniques for Graph Convolutional Networks, comparing gradient-based and decomposition-based methods on toy and chemistry datasets to facilitate understanding of their decision processes.

Contribution

It provides a comparative analysis of explainability methods for GCNs and establishes a foundation for future research and real-world applications.

Findings

Gradient-based and decomposition-based methods have different strengths.

Explainability techniques can be effectively applied to chemistry datasets.

The study highlights the need for further development of explainability tools.

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.