How Faithful are Self-Explainable GNNs?

TL;DR

This paper evaluates the faithfulness of self-explainable GNNs, revealing limitations in their interpretability and measurement methods, and discusses future directions for improving their reliability.

Contribution

It provides an analysis of faithfulness in self-explainable GNNs, highlighting current limitations and proposing avenues for enhancing their interpretability.

Findings

Identified limitations in faithfulness of current self-explainable GNNs

Highlighted issues with evaluation metrics for faithfulness

Suggested potential improvements for model and metric design

Abstract

Self-explainable deep neural networks are a recent class of models that can output ante-hoc local explanations that are faithful to the model's reasoning, and as such represent a step forward toward filling the gap between expressiveness and interpretability. Self-explainable graph neural networks (GNNs) aim at achieving the same in the context of graph data. This begs the question: do these models fulfill their implicit guarantees in terms of faithfulness? In this extended abstract, we analyze the faithfulness of several self-explainable GNNs using different measures of faithfulness, identify several limitations -- both in the models themselves and in the evaluation metrics -- and outline possible ways forward.