On the human-recognizability phenomenon of adversarially trained deep image classifiers

1 code implementation • 18 Dec 2020 • Jonathan Helland, Nathan VanHoudnos

In this work, we investigate the phenomenon that robust image classifiers have human-recognizable features -- often referred to as interpretability -- as revealed through the input gradients of their score functions and their subsequent adversarial perturbations.

Adversarial Robustness