Defending Against Physically Realizable Attacks on Image Classification
This addresses the security vulnerability of image classification systems in real-world applications, offering the first effective generic defense against physically realizable attacks.
The paper tackled the problem of defending image classification models against physically realizable attacks, showing that existing methods like adversarial training and randomized smoothing are ineffective, and proposed a new rectangular occlusion attack that, when used in adversarial training, yields models with high robustness against such attacks.
We study the problem of defending deep neural network approaches for image classification from physically realizable attacks. First, we demonstrate that the two most scalable and effective methods for learning robust models, adversarial training with PGD attacks and randomized smoothing, exhibit very limited effectiveness against three of the highest profile physical attacks. Next, we propose a new abstract adversarial model, rectangular occlusion attacks, in which an adversary places a small adversarially crafted rectangle in an image, and develop two approaches for efficiently computing the resulting adversarial examples. Finally, we demonstrate that adversarial training using our new attack yields image classification models that exhibit high robustness against the physically realizable attacks we study, offering the first effective generic defense against such attacks.