AI Paper Atlas

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

Improving Generalizability and Undetectability for Targeted Adversarial Attacks on Multimodal Pre-trained Models

Zhifang Zhang, Jiahan Zhang, Shengjie Zhou, Qi Wei, Shuo He, Feng Liu, Lei Feng

TPAMI 2026 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transfer attacks on face recognition · 8 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 single-target alignment with proxy targets plus source-modal proxiesIllusion Attack · Classification (XmediaNet+ImageNet), no defenseCls ASR (%) avg: 93.18 -> 98.31Improved
add source-modal proxy loss and target proxies under anomaly detectionIllusion Attack · Classification (XmediaNet+ImageNet), with anomaly detectionCls ASRD (%) avg: 20.04 -> 84.17Improved
proxy-based targetsIllusion Attack · Retrieval (XmediaNet+MSCOCO), no defenseR@1 ASR (%) avg: 46.49 -> 77.72Improved
proxy-based targets plus source proxiesIllusion Attack · Retrieval (XmediaNet+MSCOCO), with anomaly detectionR@1 ASRD (%) avg: 14.84 -> 61.80Improved
use PTA instead of Illusion Attack for poisoning a retrieval galleryIllusion Attack · MSCOCO retrieval poisoning, ImageBindR@1 (%) after injection: 37.42 -> 20.04 (PTA 1% vs Illusion 1%)Improved
use PTA at 0.1% injection vs Illusion Attack at 1%Illusion Attack · MSCOCO retrieval poisoning, One-PEACER@1 (%) after injection: 36.55 vs 37.26No clear effect
add more proxies via higher injection ratio of PTA AEs (ablation)PTA · MSCOCO retrieval poisoning, LanguageBindR@1 (%) after injection: 36.23 -> 15.47 -> 12.26 (0.1%, 0.5%, 1%)Improved
PTA instead of Illusion Attack under query-based black-box attackIllusion Attack · MSCOCO retrieval, ImageBind, 2x10^4 queriesR@1 ASR (%): 11.56 -> 41.81Improved
PTA instead of Illusion Attack for transfer black-box attackIllusion Attack · ImageNet zero-shot classification, ImageBind -> OpenCLIP-ViTCls ASR (%): 28.65 -> 39.19Improved
PTA instead of Illusion Attack for transfer attack OpenCLIP-ViT -> ImageBindIllusion Attack · ImageNet zero-shot classificationCls ASR (%): 30.27 -> 30.81No clear effect
PTA for textual AEsIllusion Attack · Textual AEs, classification, no defenseCls ASR (%): 26.01 -> 37.32Improved
PTA for audio AEs under defenseIllusion Attack · Audio AEs, classification, with defenseCls ASRD (%): 21.45 -> 91.03Improved

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