AI Paper Atlas

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

Meta Gradient Adversarial Attack

Zheng Yuan, Jie Zhang, Yunpei Jia, Chuanqi Tan, Tao Xue, Shiguang Shan

ICCV 2021 · Adversarial robustness and security › Black-box and transfer attacks · approach: Query-efficient black-box attacks · 101 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 MGAA to DIM (targeted)DIM · ImageNet targeted, black-box avgtargeted attack success rate (%): 19.42 -> 47.90Improved
add MGAA to MIM (targeted)MIM · ImageNet targeted, black-box avgtargeted attack success rate (%): 2.84 -> 8.56Improved
add MGAA to TI-DIM (targeted)TI-DIM · ImageNet targeted, black-box avgtargeted attack success rate (%): 23.86 -> 38.18Improved
add MGAA to TI-DIM (targeted), white-boxTI-DIM · ImageNet targeted, white-box avgtargeted attack success rate (%): 42.85 -> 85.52Improved
add MGAA to DIM (untargeted)DIM · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 79.4 -> 88.0Improved
add MGAA to MIM (untargeted)MIM · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 44.4 -> 52.0Improved
add MGAA to TI-DIM on CIFAR10TI-DIM · CIFAR10 untargeted, Adv_ResNet-18_lluntargeted attack success rate (%): 46.69 -> 68.77Improved
add meta-train ensemble step (ablation)MGAA · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 81.5 -> 99.1Improved
add meta-test step (ablation)MGAA · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 96.3 -> 99.1Improved
increase meta-train iterations K from 1 to 8 (ablation)MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 96.3 -> 99.1Improved
increase sampled tasks T from 10 to 120 (ablation)MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 94.5 -> 98.2Improved
increase ensembled models n beyond 5 (ablation)MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3untargeted attack success rate (%): 98.6 -> 98.6No clear effect

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