AdvDrop: Adversarial Attack to DNNs by Dropping Information
This work addresses a vulnerability in DNNs for computer vision applications, highlighting a novel attack perspective that could impact security-critical systems, though it is incremental in exploring adversarial robustness from a different angle.
The paper tackles the problem of Deep Neural Networks (DNNs) struggling to recognize abstract objects with lost information by proposing AdvDrop, an adversarial attack that crafts examples by dropping imperceptible details from images, resulting in a 20-30% drop in accuracy on standard datasets like ImageNet and CIFAR-10, which is more difficult to defend against than previous attacks.
Human can easily recognize visual objects with lost information: even losing most details with only contour reserved, e.g. cartoon. However, in terms of visual perception of Deep Neural Networks (DNNs), the ability for recognizing abstract objects (visual objects with lost information) is still a challenge. In this work, we investigate this issue from an adversarial viewpoint: will the performance of DNNs decrease even for the images only losing a little information? Towards this end, we propose a novel adversarial attack, named \textit{AdvDrop}, which crafts adversarial examples by dropping existing information of images. Previously, most adversarial attacks add extra disturbing information on clean images explicitly. Opposite to previous works, our proposed work explores the adversarial robustness of DNN models in a novel perspective by dropping imperceptible details to craft adversarial examples. We demonstrate the effectiveness of \textit{AdvDrop} by extensive experiments, and show that this new type of adversarial examples is more difficult to be defended by current defense systems.