Research map / Vision-language and multimodal / Human and embodied modeling
Attacking Gray-Box Large Vision-Language Models with Adaptive SVD-Structured Adversarial Alignment
Daizong Liu, Xiaowen Cai, Junhao Dong, Zhongliang Guo, Xiaoye Qu, Runwei Guan, Xiang Fang, Dengpan Ye
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add coarse alignment to the no-alignment baseline (ablation) | Qwen2.5-VL-3B attack · COCO | ASR: 0.0 -> 69.8 | Improved |
| add fine-grained alignment to the no-alignment baseline (ablation) | Qwen2.5-VL-3B attack · COCO | ASR: 0.0 -> 78.5 | Improved |
| add fine-grained alignment on top of coarse alignment (ablation) | Qwen2.5-VL-3B attack · COCO | ASR: 69.8 -> 86.5 | Improved |
| add coarse alignment on top of fine-grained alignment (ablation) | Qwen2.5-VL-3B attack · COCO | ASR: 78.5 -> 86.5 | Improved |
| add coarse alignment on top of fine-grained alignment (ablation) | LLaVa-1.5-7B attack · COCO | ASR: 75.2 -> 84.7 | Improved |
| add fine-grained alignment on top of coarse alignment (ablation) | LLaVa-1.5-7B attack · COCO | ASR: 63.7 -> 84.7 | Improved |
| use projection-based cost C_Pi in OT (ablation) | Ours · COCO, Qwen2.5-VL-3B | ASR: 82.4 -> 86.5 | Improved |
| use projection-based cost C_Pi in OT (ablation) | Ours · COCO, Gemma-3-12B | ASR: 71.6 -> 74.2 | Improved |
| use projection-based cost C_Pi in OT (ablation) | Ours · COCO, LLaVa-1.6-7B | ASR: 75.1 -> 78.5 | Improved |
| replace CLIP-ViT text encoder with Sentence-Bert (ablation) | Ours · COCO, Qwen2.5-VL-3B | ASR: 86.5 -> 87.3 | No clear effect |
| replace CLIP-ViT text encoder with Sentence-Bert (ablation) | Ours · COCO, LLaVa-1.5-7B | ASR: 84.7 -> 85.1 | No clear effect |
| increase number K of principal components in projection (ablation) | Ours · COCO, Qwen2.5-VL-3B and LLaVa-1.5-7B | similarity | Improved |
Closest papers by meaning
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- 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)
- Black-Box Membership Inference Attack for LVLMs via Prior Knowledge-Calibrated Memory Probing (NeurIPS 2025 · similarity 0.90)
- SLADE: Shielding against Dual Exploits in Large Vision-Language Models (CVPR 2025 · similarity 0.89)
- 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.
