Post-hoc gradient-based interpretability methods [1, 2] that provide instancespecific explanations of model predictions are often based on assumption (A): magnitude of input gradients--gradients of logits with respect to input--noisily highlight discriminative task-relevant features. In this work, we test the validity of assumption (A) using a three-pronged approach: 1. We develop an evaluation framework, DiffROAR, to test assumption (A) on four image classification benchmarks. Our results suggest that (i) input gradients of standard models (i.e., trained on original data) may grossly violate (A), whereas (ii) input gradients of adversarially robust models satisfy (A) reasonably well.

Yiwen Guo leads the project and serves as the corresponding author.

We first demonstrate theoretically that off-manifold robustness leads input gradients to lie approximately on the data manifold, explaining their perceptual alignment. We then show that Bayes optimal models satisfy off-manifold robustness, and confirm the same empirically for robust models trained via gradient norm regularization, randomized smoothing, and adversarial training with projected gradient descent. Quantifying the perceptual alignment of model gradients via their similarity with the gradients of generative models, we show that off-manifold robustness correlates well with perceptual alignment. Finally, based on the levels of on-and off-manifold robustness, we identify three different regimes of robustness that affect both perceptual alignment and model accuracy: weak robustness, bayes-aligned robustness, and excessive robustness.

Recent approaches include priors on the feature attribution of a deep neural network (DNN) into the training process to reduce the dependence on unwanted features. However, until now one needed to trade off high-quality attributions, satisfying desirable axioms, against the time required to compute them. This in turn either led to long training times or ineffective attribution priors.

The null block, containing the square patch, does not encode any information of the image class.

The null block, containing the square patch, does not encode any information of the image class.