Improving Stability Estimates in Adversarial Explainable AI through Alternate Search Methods

TL;DR

This paper introduces an alternative search method to better quantify the stability of local surrogate explanations in adversarial settings, highlighting the importance of minimal perturbations for assessing explanation robustness.

Contribution

It proposes a novel search approach to measure the minimal perturbations needed to alter explanations, enabling more accurate comparison of explainability method stability.

Findings

The new method effectively identifies smaller perturbations than previous approaches.

It provides a nuanced measure of explanation stability based on minimal perturbation requirements.

The approach improves the evaluation of robustness in adversarial explainability scenarios.

Abstract

Advances in the effectiveness of machine learning models have come at the cost of enormous complexity resulting in a poor understanding of how they function. Local surrogate methods have been used to approximate the workings of these complex models, but recent work has revealed their vulnerability to adversarial attacks where the explanation produced is appreciably different while the meaning and structure of the complex model's output remains similar. This prior work has focused on the existence of these weaknesses but not on their magnitude. Here we explore using an alternate search method with the goal of finding minimum viable perturbations, the fewest perturbations necessary to achieve a fixed similarity value between the original and altered text's explanation. Intuitively, a method that requires fewer perturbations to expose a given level of instability is inferior to one which requires more. This nuance allows for superior comparisons of the stability of explainability methods.