Carpet-bombing patch: attacking a deep network without usual requirements

TL;DR

This paper introduces a novel patch attack method called carpet-bombing that effectively reduces deep network performance across various tasks without requiring task-specific knowledge or constraints.

Contribution

It presents a new universal patch attack that targets feature representations, removing the need for task-specific information or assumptions.

Findings

Decreases accuracy on ImageNet

Reduces mAP on Pascal VOC

Lowers IoU on Cityscapes

Abstract

Although deep networks have shown vulnerability to evasion attacks, such attacks have usually unrealistic requirements. Recent literature discussed the possibility to remove or not some of these requirements. This paper contributes to this literature by introducing a carpet-bombing patch attack which has almost no requirement. Targeting the feature representations, this patch attack does not require knowing the network task. This attack decreases accuracy on Imagenet, mAP on Pascal Voc, and IoU on Cityscapes without being aware that the underlying tasks involved classification, detection or semantic segmentation, respectively. Beyond the potential safety issues raised by this attack, the impact of the carpet-bombing attack highlights some interesting property of deep network layer dynamic.