Research map / Adversarial robustness and security / Attacks on text and graphs
Transferable and Stealthy Adversarial Attacks on Large Vision-Language Models
Zhewen Yao, Yao Zhu, Shiliang Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add progressive alignment objectives (ablation) | PSI · GPT-5 | ASR: 22.8 -> 78.6 | Improved |
| add progressive alignment objectives (ablation) | PSI · GPT-5 | S-ASR: 15.0 -> 62.8 | Improved |
| add diffusion framework instead of l_inf pixel attack (16/255) (ablation) | PSI · GPT-5 | BRISQUE: 51.49 -> 22.14 | Improved |
| add diffusion framework instead of l_inf pixel attack (16/255) (ablation) | PSI · GPT-5 | S-ASR: 57.0 -> 62.8 | Improved |
| add co-evolving region selection (ablation) | PSI · GPT-5 | ASR: 71.3 -> 78.6 | Improved |
| add co-evolving region selection (ablation) | PSI · GPT-5 | BRISQUE: 25.60 -> 22.14 | Improved |
| add source-aware denoising (ablation) | PSI · GPT-5 | LPIPS: 0.241 -> 0.192 | Improved |
| add source-aware denoising (ablation) | PSI · GPT-5 | ASR: 81.0 -> 78.6 | Hurt |
| spread perturbation from timestep 0.2 to 0 instead of only at 0.2 (ablation) | PSI · GPT-5 | ASR: 57.2% -> 78.6% | Improved |
| select most similar source region instead of random (ablation) | PSI · global similarity @100 | global similarity: 0.67 -> 0.75 | Improved |
| use cropped dominant object as target instead of full target image (ablation) | M-Attack · GPT-5 | ASR: 73.8% -> 81.4% | Improved |
| use PSI instead of FOA against DiffPure defense | FOA · GPT-5 under DiffPure | ASR: 19.7 -> 34.2 | Improved |
Closest papers by meaning
- A Frustratingly Simple Yet Highly Effective Attack Baseline: Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1 (NeurIPS 2025 · similarity 0.90)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- SLADE: Shielding against Dual Exploits in Large Vision-Language Models (CVPR 2025 · similarity 0.89)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.88)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.88)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (ICLR 2025 · similarity 0.88)
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.
