AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

MTAttack: Multi-Target Backdoor Attacks Against Large Vision-Language Models

Zihan Wang, Guansong Pang, Wenjun Miao, Jin Zheng, Xiao Bai

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 3 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 TPA constraint on top of PSP (ablation)MTAttack · COCO 4-target, LLaVA-1.5-7bTCR: 1.23 -> 0.58Improved
add TPA constraint on top of PSP (ablation)MTAttack · COCO 4-target, LLaVA-1.5-7bCIDEr: 127.82 -> 129.37Improved
add PSP constraint on top of TPA (ablation)MTAttack · COCO 4-target, LLaVA-1.5-7bASR: 98.48 -> 99.43Improved
use MTAttack instead of VL-Trojan for 4-target attackQwen2.5-VL-7b · Flickr-30K 4-targetTCR: 46.20 -> 4.25Improved
bind triggers to proxy classes rather than pulling toward target classMTAttack · Flickr-30K 4-target, MiniGPT-v2ASR: 75.83 -> 99.03Improved
use more triggers (4 vs 1) with baseline methodsSIG, ReFool, WaNet baselines · Qwen2.5-VL clean-image qualityGPT scoreHurt
optimize triggers on Flickr-30K and test on COCOMTAttack · Cross-dataset, MiniGPT-v2 4-targetASR: 99.03 -> 99.93No clear effect
change attack concept after trigger optimizationMTAttack · Qwen2.5-VL Flickr-30K, 4-target road signASR: 95.55 -> 96.80No clear effect
exclude 20% most suspicious samples via detectorMTAttack · Flickr-30K 4-target, LLaVA-1.5ASR: 98.13 -> 96.80No clear effect
exclude 20% most suspicious samples via detectorVL-Trojan · Flickr-30K 4-target, LLaVA-1.5ASR: 92.18 -> 63.70Hurt
add JPEG compression defenseVL-Trojan · Flickr-30K 4-target, LLaVA-1.5ASR: 92.18 -> 35.10Hurt
add JPEG compression defense with augmentation during trigger optimizationMTAttack · Flickr-30K 4-target, LLaVA-1.5ASR: 98.13 -> 96.43No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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