Generalizing universal adversarial perturbations for deep neural networks

Previous studies have shown that universal adversarial attacks can fool deep neural networks over a large set of input images with a single human-invisible perturbation. However, current methods for universal adversarial attacks are based on additive perturbation, which enables misclassification by directly adding the perturbation on the input images. In this paper, for the first time, we show that a universal adversarial attack can also be achieved through spatial transformation (non-additive). More importantly, to unify both additive and non-additive perturbations, we propose a novel unified yet flexible framework for universal adversarial attacks, called GUAP, which can initiate attacks by l(8)-norm (additive) perturbation, spatially-transformed (non-additive) perturbation, or a combination of both. Extensive experiments are conducted on two computer vision scenarios, including image classification and semantic segmentation tasks, which contain CIFAR-10, ImageNet and Cityscapes datasets with a number of different deep neural network models, including GoogLeNet, VGG16/19, ResNet101/152, DenseNet121, and FCN-8s. Empirical experiments demonstrate that GUAP can obtain higher attack success rates on these datasets compared to state-of-the-art universal adversarial attacks. In addition, we also demonstrate how universal adversarial training benefits the robustness of the model against universal attacks. We release our tool GUAP on https://github.com/TrustAI/GUAP.

Automated summary

Previous studies have shown that deep neural networks can be fooled by universal adversarial attacks using a single human-invisible perturbation. This paper introduces a novel unified framework called GUAP, which enables both additive and non-additive perturbations for universal adversarial attacks. Extensive experiments demonstrate that GUAP outperforms state-of-the-art methods in terms of attack success rates on various datasets and computer vision tasks.