Provable Robust Saliency-based Explanations

TL;DR

This paper introduces R2ET, a novel training method that enhances the stability and robustness of saliency-based explanations in machine learning models, addressing limitations of existing metrics and adversarial training.

Contribution

The paper proposes a new stability metric, R2ET training method, and theoretical analysis linking it to certified robustness for saliency explanations.

Findings

R2ET improves explanation stability against stealthy attacks

R2ET generalizes across various explanation methods and data modalities

Theoretical analysis confirms R2ET's robustness guarantees

Abstract

To foster trust in machine learning models, explanations must be faithful and stable for consistent insights. Existing relevant works rely on the $\ell_p$ distance for stability assessment, which diverges from human perception. Besides, existing adversarial training (AT) associated with intensive computations may lead to an arms race. To address these challenges, we introduce a novel metric to assess the stability of top-$k$ salient features. We introduce R2ET which trains for stable explanation by efficient and effective regularizer, and analyze R2ET by multi-objective optimization to prove numerical and statistical stability of explanations. Moreover, theoretical connections between R2ET and certified robustness justify R2ET's stability in all attacks. Extensive experiments across various data modalities and model architectures show that R2ET achieves superior stability against stealthy attacks, and generalizes effectively across different explanation methods.