Research map / Adversarial robustness and security / Attacks on text and graphs
Fooling Vision and Language Models Despite Localization and Attention Mechanism
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the proposed VQA targeted attack instead of CW | MCB on VQA-A · VQA-A | attack success rate (%): 94.10 -> 94.67; 94.28 -> 94.78; 94.27 -> 94.97; 94.52 -> 95.02; 94.78 -> 95.15 | Helped |
| Use the proposed VQA targeted attack instead of CW | N2NMN on VQA-A · VQA-A | attack success rate (%): 93.82 -> 94.25; 93.78 -> 94.53; 95.02 -> 95.57; 95.08 -> 95.80; 95.37 -> 96.15 | Helped |
| Use the proposed VQA attack instead of CW | MCB and N2NMN · VQA-A | adversarial probability | Helped |
| Use question-answer targets that are semantically compatible rather than question-incompatible | VQA targeted attacks · VQA | attack success rate: 7.8%; 4.6% (question-incompatible targets) | Helped |
| Attack question-incompatible target answers | MCB and N2NMN · Non-Sense | attack success rate: 7.8%; 4.6% | Hurt |
| Use N2NMN rather than MCB as the victim model | VQA targeted attacks · VQA-A | adversarial probability | Hurt |
| Attack the caption generator while keeping proposed regions unchanged | DenseCap · Visual Genome | attack success rate: 99.65% | Helped |
| Use targeted adversarial examples on dense captioning | DenseCap · Caption B | attack success rate: 97.24% | Helped |
Closest papers by meaning
- Visual Adversarial Examples Jailbreak Aligned Large Language Models (AAAI 2024 · similarity 0.89)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.89)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.88)
- Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models (CVPR 2025 · similarity 0.88)
- One Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models (CVPR 2024 · 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.