On the Benefits of Models with Perceptually-Aligned Gradients

Adversarial robust models have been shown to learn more robust and interpretable features than standard trained models. As shown in [citetsipras2018robustness], such robust models inherit useful interpretable properties where the gradient aligns perceptually well with images, and adding a large targeted adversarial perturbation leads to an image resembling the target class. We perform experiments to show that interpretable and perceptually aligned gradients are present even in models that do not show high robustness to adversarial attacks. Specifically, we perform adversarial training with attack for different max-perturbation bound. Adversarial training with low max-perturbation bound results in models that have interpretable features with only slight drop in performance over clean samples. In this paper, we leverage models with interpretable perceptually-aligned features and show that adversarial training with low max-perturbation bound can improve the performance of models for zero-shot and weakly supervised localization tasks.