AI Paper Atlas

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

Test-Time Attention Purification for Backdoored Large Vision Language Models

Zhifang Zhang, Bojun Yang, Shuo He, Weitong Chen, Wei Emma Zhang, Olaf Maennel, Lei Feng, Miao Xu

CVPR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 2 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 CleanSight attention-based detection and pruning instead of pixel-space purificationbackdoored LLaVA-1.5-7B · VQAv2 (BadNet, adapter)ASR (%): 100.00 -> 0Improved
use CleanSight instead of ZIP across attacksbackdoored LLaVA-1.5-7B · VQAv2 (Blended)ASR (%): 84.77 -> 0Improved
add CleanSight on clean inputsbackdoored LLaVA-1.5-7B · VQAv2 (BadNet), clean utilityV-score: 62.89 -> 62.63No clear effect
add attention-stealing detection before pruning (instead of pruning all inputs) (ablation)CleanSight · VQAv2 (BadNet), clean utilityV-score: 52.47 -> 62.63Improved
use attention-guided selective pruning instead of random 90% visual token removal (ablation)CleanSight · VQAv2 (BadNet)ASR (%): 100.00 -> 0Improved
apply pixel perturbation (transformation defense) to backdoored LVLMLLaVA backdoored model · Figure 1 attack success rateASRNo clear effect
flatten visual attention weights toward uniformbackdoored LLaVA · Figure 1 attack success rateASRImproved
lower pruning threshold tau to about 1e-3 (oracle detector) (ablation)CleanSight · VQAv2 (four attacks)ASR (%): ASR drops to nearly zeroImproved
make tau overly small (ablation)CleanSight · VQAv2 poisoned utilityV-score on poisoned inputHurt
increase clean validation set size (ablation)CleanSight detector · BadNet and Blended detectionFPRImproved
use head-specific attention ratio instead of head-averaged ratio (ablation)CleanSight detector · AUROC across layersAUROCImproved
apply CleanSight to other LVLMs and a LoRA-tuned modelLLaVA-1.5-7B LoRA, LLaVA-1.5-13B, InstructBLIP-7B, Qwen-VL2-7B · VQAv2 (BadNet)ASR (%): LoRA 100.00 -> 5.08; Qwen-VL2 100.00 -> 0Improved

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