Scratch that! An Evolution-based Adversarial Attack against Neural Networks
This addresses security vulnerabilities in neural networks for image classification, with incremental improvements in query efficiency and applicability to real-world APIs.
The paper tackles the problem of generating black-box adversarial attacks on image classifiers by modifying a small fraction of pixels as scratches, achieving targeted success rates of up to 98.77% on ImageNet and 97.20% on CIFAR-10 with less than 5% pixel coverage.
We study black-box adversarial attacks for image classifiers in a constrained threat model, where adversaries can only modify a small fraction of pixels in the form of scratches on an image. We show that it is possible for adversaries to generate localized \textit{adversarial scratches} that cover less than $5\%$ of the pixels in an image and achieve targeted success rates of $98.77\%$ and $97.20\%$ on ImageNet and CIFAR-10 trained ResNet-50 models, respectively. We demonstrate that our scratches are effective under diverse shapes, such as straight lines or parabolic B\a'ezier curves, with single or multiple colors. In an extreme condition, in which our scratches are a single color, we obtain a targeted attack success rate of $66\%$ on CIFAR-10 with an order of magnitude fewer queries than comparable attacks. We successfully launch our attack against Microsoft's Cognitive Services Image Captioning API and propose various mitigation strategies.