Understanding Adversarial Robustness Against On-manifold Adversarial Examples. (arXiv:2210.00430v1 [cs.LG])

Deep neural networks (DNNs) are shown to be vulnerable to adversarial
examples. A well-trained model can be easily attacked by adding small
perturbations to the original data. One of the hypotheses of the existence of
the adversarial examples is the off-manifold assumption: adversarial examples
lie off the data manifold. However, recent research showed that on-manifold
adversarial examples also exist. In this paper, we revisit the off-manifold
assumption and want to study a question: at what level is the poor performance …

adversarial manifold robustness