AI Paper Atlas

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

Boosting Transferability of Targeted Adversarial Examples via Hierarchical Generative Networks

Xiao Yang, Yinpeng Dong, Tianyu Pang, Hang Su, Jun Zhu

ECCV 2022 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transferable adversarial examples · 39 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 instance-specific Logit attack with the conditional generatorLogit · NeurIPS 1k, 8 targets, Inv3 source, target Inv4targeted success rate: 5.6 -> 66.9Improved
add smooth tanh projection and adaptive Gaussian smoothing over CD-APCD-AP · NeurIPS 1k, 8 targets, Inv3 source, target Inv3ens3targeted success rate: 1.5 -> 37.2Improved
use one conditional model instead of 8 separate CD-AP modelsCD-AP · NeurIPS 1k, 8 targets, Inv3 source, target Inv4targeted success rate: 57.6 -> 66.9Improved
replace Gaussian smoothing of CD-AP-gs with the full methodCD-AP-gs · NeurIPS 1k, 8 targets, Inv3 source, target Inv4targeted success rate: 31.3 -> 66.9Improved
hierarchical k-DPP partition into 20 models of 50 classesConditional generator · NeurIPS 2017 competition, 1000 target classes, target Inv4targeted success rate: 4.7 -> 35.9Improved
reduce perturbation budget from eps 16 to eps 8 (ablation)Ours · Targeted black-box, Inv3 source, target VGG-16targeted success rate: 61.9 -> 36.1Hurt
reduce eps from 16 to 8 on R152 target (ablation)Ours · Targeted black-box, Inv3 source, target R152targeted success rate: 49.6 -> 16.3Hurt
compare Ours against Logit at eps 8Logit · Targeted black-box, Inv3 source, target VGG-16, eps=8targeted success rate: 2.1 -> 36.1Improved
train generator on MS-COCO instead of ImageNet (ablation)Ours · Targeted black-box, Inv3 substitute, target DNtargeted success rate: 79.9 -> 70.3Hurt
train generator on Comics instead of ImageNet (ablation)Ours · Targeted black-box, Inv3 substitute, target DNtargeted success rate: 79.9 -> 60.4Hurt
use diverse conditional classes from k-DPP instead of cosine-close classes (ablation)Ours · ImageNet target class subsetsloss convergence and transferabilityImproved
increase the number of target classes handled by a single generator (ablation)Ours · ImageNet target classestargeted 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