AI Paper Atlas

Research map / Adversarial robustness and security / Black-box and transfer attacks

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

NeurIPS 2022 · Adversarial robustness and security › Black-box and transfer attacks · approach: Query-efficient black-box attacks · 10 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add AAA post-processing defenseWideResNet-28 · CIFAR-10 Square 2500 queriesadversarial accuracy %: 00.09 -> 80.59Improved
AAA instead of adversarial trainingWideResNet-28 · CIFAR-10 Square 2500 queriesadversarial accuracy %: 67.44 -> 80.59Improved
add AAAWideResNet-28 · CIFAR-10 cleanaccuracy %: 94.78 -> 94.84No clear effect
add AAAWideResNet-28 · CIFAR-10 cleanECE %: 3.52 -> 2.46Improved
add adversarial trainingWideResNet-28 · CIFAR-10 cleanaccuracy %: 94.78 -> 87.02Hurt
add random noise defense RNDWideResNet-28 · CIFAR-10 cleanECE %: 3.52 -> 6.32Hurt
combine AAA with adversarial training (ablation)AT model · CIFAR-10 Square 2500 queries l2=2.5adversarial accuracy %: 25.19 -> 73.72Improved
combine AAA with adversarial training (ablation)AT model · CIFAR-10 cleanECE %: 11.00 -> 10.56No clear effect
increase AAA optimization iterations from 0 to 20 (ablation)AAA-linear · CIFAR-10 Square 100 queriesadversarial accuracy %: 39.38 -> 79.29Improved
increase iterations from 60 to 100 (ablation)AAA-linear · CIFAR-10 Square 100 queriesadversarial accuracy %: 81.37 -> 81.36No clear effect
adaptive opposite-direction attacker against linear designAAA-linear · CIFAR-10 Square 100 queriesadversarial accuracy %: 81.36 -> 57.31Hurt
use sine target loss instead of linear (ablation)AAA · CIFAR-10 adaptive Square 100 queriesadversarial accuracy %: 57.31 -> 76.41Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy