Stochastic sparse adversarial attacks

TL;DR

This paper presents stochastic sparse adversarial attacks (SSAA), a fast, noise-based method for generating sparse adversarial examples for neural networks, demonstrating superior efficiency and effectiveness on large datasets like ImageNet.

Contribution

Introduces SSAA, a novel noise-based sparse attack method utilizing small-time expansion, with scalable and more effective results compared to existing approaches.

Findings

VFGA scales efficiently to ImageNet

VFGA achieves lower L0 scores than SparseFool

VFGA outperforms Sparse-RS in success rate and sparsity

Abstract

This paper introduces stochastic sparse adversarial attacks (SSAA), standing as simple, fast and purely noise-based targeted and untargeted attacks of neural network classifiers (NNC). SSAA offer new examples of sparse (or $L_0$) attacks for which only few methods have been proposed previously. These attacks are devised by exploiting a small-time expansion idea widely used for Markov processes. Experiments on small and large datasets (CIFAR-10 and ImageNet) illustrate several advantages of SSAA in comparison with the-state-of-the-art methods. For instance, in the untargeted case, our method called Voting Folded Gaussian Attack (VFGA) scales efficiently to ImageNet and achieves a significantly lower $L_0$ score than SparseFool (up to $\frac{2}{5}$) while being faster. Moreover, VFGA achieves better $L_0$ scores on ImageNet than Sparse-RS when both attacks are fully successful on a large number of samples.