AIM: Adversarial Information Masking for Faithfulness Evaluation of Saliency Maps

TL;DR

AIM introduces an adversarial feature replacement framework to evaluate the faithfulness of saliency maps, addressing biases caused by traditional masking methods across multiple modalities.

Contribution

The paper proposes a novel adversarial masking approach that improves the reliability of saliency map evaluations by reducing masking-induced bias.

Findings

AIM reduces bias compared to zero and interpolation masking methods.

The method reveals modality-dependent differences in attribution methods.

Experiments across image, audio, and EEG data validate AIM's effectiveness.

Abstract

Post-hoc saliency methods are widely used to interpret deep neural networks, but their faithfulness is difficult to evaluate reliably. Existing evaluations mask features according to saliency-induced feature ordering and measure performance degradation, but this degradation can be confounded by the masking operator: zero masking may create out-of-distribution artifacts, while interpolation-based masking may preserve residual predictive information. We propose Adversarial Information Masking (AIM), a saliency-guided adversarial feature replacement framework for evaluating both saliency-map faithfulness and masking-operator reliability. AIM replaces selected features with values from an adversarial counterpart of the input and compares degradation under complementary masking orders. We assess reliability using random-attribution bias and stability of explanation-method faithfulness rankings. Experiments on image, audio, and EEG tasks suggest that AIM reduces masking-induced bias compared with zero and interpolation-based masking, while revealing modality-dependent differences between signed and unsigned attributions.