Multiple Perturbation Attack: Attack Pixelwise Under Different $\ell_p$-norms For Better Adversarial Performance
Preprint
Abstract
Adversarial machine learning has been both a major concern and a hot topic recently, especially with the ubiquitous use of deep neural networks in the current landscape. Adversarial attacks and defenses are usually likened to a cat-and-mouse game in which defenders and attackers evolve over the time. On one hand, the goal is to develop strong and robust deep networks that are resistant to malicious actors. On the other hand, in order to achieve that, we need to devise even stronger adversarial attacks to challenge these defense models. Most of existing attacks employs a single $\ell_p$ distance (commonly, $p\in\{1,2,\infty\}$) to define the concept of closeness and performs steepest gradient ascent w.r.t. this $p$-norm to update all pixels in an adversarial example in the same way. These $\ell_p$ attacks each has its own pros and cons; and there is no single attack that can successfully break through defense models that are robust against multiple $\ell_p$ norms simultaneously. Motivated by these observations, we come up with a natural approach: combining various $\ell_p$ gradient projections on a pixel level to achieve a joint adversarial perturbation. Specifically, we learn how to perturb each pixel to maximize the attack performance, while maintaining the overall visual imperceptibility of adversarial examples. Finally, through various experiments with standardized benchmarks, we show that our method outperforms most current strong attacks across state-of-the-art defense mechanisms, while retaining its ability to remain clean visually.
Keywords
All Related Versions
This publication has 9 references indexed in Scilit:
- A Light Recipe to Train Robust Vision Transformers2023
- Deep face recognition: A surveyNeurocomputing, 2021
- A Discussion of 'Adversarial Examples Are Not Bugs, They Are Features'Distill, 2019
- Revisiting Normalized Gradient Descent: Fast Evasion of Saddle PointsIEEE Transactions on Automatic Control, 2019
- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric2018
- Towards Evaluating the Robustness of Neural Networks2017
- Review of advanced driver assistance systems (ADAS)AIP Conference Proceedings, 2017
- DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks2016
- Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification2015