AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

MADA-Attack: Transferable Multi-modal Attention Distraction Adversarial Attack against Vision Language Models

Zhihan Qin, Jiahao Chen, Chunyi Zhou, Yuwen Pu, Chunqiang Hu, Xiaolei Liu, Shouling Ji

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Contrastive vision-language alignment

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 Semantic Token Manipulation (structured text perturbation) instead of random token selection (ablation)MADA-Attack · MSCOCO captioning, BLIP2CIDEr ASR (%): 45.63 -> 49.42Improved
add Fused Embedding Training (joint text and image optimization) instead of image-only (ablation)MADA-Attack · MSCOCO captioning, BLIP2CIDEr ASR (%): 38.99 -> 49.42Improved
add Adaptive Data Augmentation (low early probability, increasing later) instead of fixed probability (ablation)MADA-Attack · MSCOCO captioning, BLIP2CIDEr ASR (%): 36.51 -> 49.42Improved
add text-guided embedding-level optimization to UAP for zero-shot classificationXTransfer Base · ZS classification average (BLIP2)ASR (%): 70.83 -> 82.60Improved
MADA-Attack vs best baseline on BLIP2 captioningXTransfer Base · Flickr30K captioning, BLIP2CIDEr ASR (%): 33.3 -> 44.39Improved
use MADA-Attack on commercial closed-source modelsC-PGC and XTransfer Naive · GPT-4o, Gemini, Claude captioningCIDEr ASR (%): GPT-4o 74.50 -> 94.59Improved
MADA-Attack on VizWiz VQA vs best baselineMiniGPT4 VQA · VizWizASR (%)No clear effect
MADA-Attack vs XTransfer Base on LLaVA-7B captioning CIDErXTransfer Base · Flickr30K captioning, LLaVA-7BCIDEr ASR (%): 97.05 -> 98.25No clear effect
increase image perturbation budget from 4/255 to 12/255 (ablation)MADA-Attack · BLIP2 captioningcaption disruption (qualitative)Improved
MADA-Attack training time vs other UAP methodsUAP training · ZS classificationtraining timeNo clear effect

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