AI Paper Atlas

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

On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression

Xinwei Zhang, Hangcheng Liu, Li Bai, Hao Wang, Qingqing Ye, Tianwei Zhang, Haibo Hu

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 0 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 compression-aligned optimization (CAGE) over baseline attackVEAttack on compressed LLaVA · TextVQA (K_model=64, avg)robust accuracy (lower is stronger): 31.1 -> 18.9Improved
use CAGE instead of baseline attackVEAttack on compressed LLaVA · VQA-v2 (K_model=16, avg)robust accuracy (lower is stronger): 50.3 -> 42.9Improved
use CAGE instead of baseline attackVEAttack on compressed LLaVA · GQA (K_model=192, avg)robust accuracy (lower is stronger): 40.8 -> 36.2Improved
add the second objective (lambda) on top of EFD only (ablation)CAGE · VQA-v2 (K_model=64)robust accuracy: 45.8 -> 43.0Improved
add the second objective on top of EFD only (ablation)CAGE · VQA-v2 (full tokens)robust accuracy: 50.8 -> 49.4Improved
use attention score instead of l1-norm score for token importance (ablation)CAGE · VQA-v2 (K_model=64)robust accuracy: 60.6 -> 43.0Improved
use attention score instead of l1-norm score (ablation)CAGE · VQA-v2 (K_model=16)robust accuracy: 51.0 -> 42.7Improved
align attack budget with model token budgetVEAttack on LLaVA · K_model=16, K_attack 576 vs 16robust accuracy: 49.7 -> 44.4Improved
increase perturbation budget from 2/255 to 8/255 (ablation)baseline attack · VQA-v2 (K_model=64)robust accuracy: 53.5 -> 43.2Improved
increase perturbation budget from 2/255 to 8/255 (ablation)CAGE · VQA-v2 (K_model=192)robust accuracy: 46.5 -> 42.6No clear effect
apply D1 robustness-aware selection as defenseVisionZip · VQA-v2 (K_model=128)robust accuracy under CAGE: 45.1 -> 50.0Improved
apply D1 defenseVisionZip · VQA-v2 (K_model=128)clean accuracy: 72.1 -> 70.9Hurt

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