How Good is your Explanation? Algorithmic Stability Measures to Assess the Quality of Explanations for Deep Neural Networks

TL;DR

This paper introduces two novel algorithmic stability measures, MeGe and ReCo, to objectively evaluate the quality of explanations for deep neural networks, highlighting their effectiveness over traditional fidelity measures.

Contribution

The paper proposes two new stability-based metrics for assessing explanation quality and demonstrates their advantages through extensive experiments on various neural network architectures and datasets.

Findings

MeGe and ReCo outperform traditional fidelity measures in evaluating explanations.

1-Lipschitz networks generate more stable and trustworthy explanations.

Explanations from 1-Lipschitz networks are comparable in accuracy to standard networks.

Abstract

A plethora of methods have been proposed to explain how deep neural networks reach their decisions but comparatively, little effort has been made to ensure that the explanations produced by these methods are objectively relevant. While several desirable properties for trustworthy explanations have been formulated, objective measures have been harder to derive. Here, we propose two new measures to evaluate explanations borrowed from the field of algorithmic stability: mean generalizability MeGe and relative consistency ReCo. We conduct extensive experiments on different network architectures, common explainability methods, and several image datasets to demonstrate the benefits of the proposed measures.In comparison to ours, popular fidelity measures are not sufficient to guarantee trustworthy explanations.Finally, we found that 1-Lipschitz networks produce explanations with higher MeGe and ReCo than common neural networks while reaching similar accuracy. This suggests that 1-Lipschitz networks are a relevant direction towards predictors that are more explainable and trustworthy.