Timezone: »

Adversarial Attack on Attackers: Post-Process to Mitigate Black-Box Score-Based Query Attacks
Sizhe Chen · Zhehao Huang · Qinghua Tao · Yingwen Wu · Cihang Xie · Xiaolin Huang

Tue Nov 29 09:00 AM -- 11:00 AM (PST) @ Hall J #708
The score-based query attacks (SQAs) pose practical threats to deep neural networks by crafting adversarial perturbations within dozens of queries, only using the model's output scores. Nonetheless, we note that if the loss trend of the outputs is slightly perturbed, SQAs could be easily misled and thereby become much less effective. Following this idea, we propose a novel defense, namely Adversarial Attack on Attackers (AAA), to confound SQAs towards incorrect attack directions by slightly modifying the output logits. In this way, (1) SQAs are prevented regardless of the model's worst-case robustness; (2) the original model predictions are hardly changed, i.e., no degradation on clean accuracy; (3) the calibration of confidence scores can be improved simultaneously. Extensive experiments are provided to verify the above advantages. For example, by setting $\ell_\infty=8/255$ on CIFAR-10, our proposed AAA helps WideResNet-28 secure 80.59% accuracy under Square attack (2500 queries), while the best prior defense (i.e., adversarial training) only attains 67.44%. Since AAA attacks SQA's general greedy strategy, such advantages of AAA over 8 defenses can be consistently observed on 8 CIFAR-10/ImageNet models under 6 SQAs, using different attack targets, bounds, norms, losses, and strategies. Moreover, AAA calibrates better without hurting the accuracy. Our code is available at https://github.com/Sizhe-Chen/AAA.

Author Information

Sizhe Chen (Shanghai Jiao Tong University)
Sizhe Chen

I’m a third-year postgraduate student at Shanghai Jiao Tong University, where I also received my B.E. degree. My research focuses on secure, robust, reliable, interpretable, and trustworthy machine learning. I’m applying for Ph.D. positions (Fall 2023).

Zhehao Huang (Shanghai Jiaotong University)
Qinghua Tao (Department of Electrical Engineering, KU Leuven, Belgium, KU Leuven)
Yingwen Wu (Shanghai Jiao Tong University)
Cihang Xie ( University of California, Santa Cruz)
Xiaolin Huang (Shanghai Jiao Tong University, Tsinghua University)

More from the Same Authors