How Faithful are Self-Explainable GNNs?

Marc Christiansen, Lea Villadsen, Zhiqiang Zhong, Stefano Teso, Davide Mottin

Keywords: deep learning, explainability, graph neural networks, self-explainable models, concepts

TL;DR: We analyze the faithfulness of self-explainable GNNs and identify limitations in the models themselves and in the evaluation metrics.

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.

Submission Type: Extended abstract (max 4 main pages).

Agreement: Check this if you are okay with being contacted to participate in an anonymous survey.

Submission Number: 88