Interpreting Global Perturbation Robustness of Image Models using Axiomatic Spectral Importance Decomposition

TL;DR

This paper introduces I-ASIDE, a spectral importance decomposition method that uses axiomatic principles to interpret the mechanisms behind image model robustness to perturbations, revealing the roles of different frequency features.

Contribution

The paper presents a novel, model-agnostic interpretability method based on spectral importance decomposition that explains perturbation robustness mechanisms in image models.

Findings

Low-frequency signals are more robust than high-frequency signals.

I-ASIDE effectively measures and interprets model robustness across various vision models.

Spectral SNR decay follows a power-law, influencing robustness understanding.

Abstract

Perturbation robustness evaluates the vulnerabilities of models, arising from a variety of perturbations, such as data corruptions and adversarial attacks. Understanding the mechanisms of perturbation robustness is critical for global interpretability. We present a model-agnostic, global mechanistic interpretability method to interpret the perturbation robustness of image models. This research is motivated by two key aspects. First, previous global interpretability works, in tandem with robustness benchmarks, e.g. mean corruption error (mCE), are not designed to directly interpret the mechanisms of perturbation robustness within image models. Second, we notice that the spectral signal-to-noise ratios (SNR) of perturbed natural images exponentially decay over the frequency. This power-law-like decay implies that: Low-frequency signals are generally more robust than high-frequency signals -- yet high classification accuracy can not be achieved by low-frequency signals alone. By applying Shapley value theory, our method axiomatically quantifies the predictive powers of robust features and non-robust features within an information theory framework. Our method, dubbed as \textbf{I-ASIDE} (\textbf{I}mage \textbf{A}xiomatic \textbf{S}pectral \textbf{I}mportance \textbf{D}ecomposition \textbf{E}xplanation), provides a unique insight into model robustness mechanisms. We conduct extensive experiments over a variety of vision models pre-trained on ImageNet to show that \textbf{I-ASIDE} can not only \textbf{measure} the perturbation robustness but also \textbf{provide interpretations} of its mechanisms.