AI Paper Atlas

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

Adaptive Perturbation for Adversarial Attack

Zheng Yuan, Jie Zhang, Zhaoyan Jiang, Liangliang Li, Shiguang Shan

TPAMI 2024 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transfer attacks on face recognition · 27 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
replace sign function with fixed scaling factor (APAA_f)TIM · ImageNet, Inc-v3attack success rate: 86.5 -> 88.6Improved
replace sign function with fixed scaling factor (APAA_f)VT · ImageNet, Inc-v3attack success rate: 89.1 -> 92.3Improved
replace sign function with fixed scaling factor (APAA_f)MIFGSM · CIFAR10, Res-18attack success rate: 88.8 -> 92.7Improved
replace sign function with fixed scaling factor (APAA_f)SIM · CIFAR10, Res-18attack success rate: 93.2 -> 96.5Improved
use adaptive scaling factor generator instead of fixed factorSIM · CIFAR10, RegNet targetattack success rate: 91.6 -> 93.8Improved
use adaptive scaling factor generator instead of fixed factorMIFGSM · CIFAR10, RegNet targetattack success rate: 85.6 -> 86.5Improved
replace sign with arctanh (ablation)MIFGSM · CIFAR10, Res-18attack success rate: 88.8 -> 55.8Hurt
replace sign with Adam (ablation)MIFGSM · CIFAR10, Res-18attack success rate: 88.8 -> 92.2Mixed
increase gamma from 0.2 to 1.0 on normal models (ablation)TIM w/ APAA_f · ImageNet, Inc-v3attack success rate: 84.6 -> 89.7Improved
increase gamma from 0.2 to 1.0 on adversarially trained models (ablation)TIM w/ APAA_f · ImageNet, Inc-v3_advattack success rate: 69.5 -> 57.1Hurt
add APAA_f and use 10 steps rather than baseline 100 stepsMIFGSM · CIFAR10, RegNet targetattack success rate: 94.4 -> 96.3Improved
train scaling generator with a single classifier (ablation)APAA_a · transferabilityattack success rateHurt

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