Research map / Adversarial robustness and security / Attacks on text and graphs
LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models
Alvi Md Ishmam, Najibul Haque Sarker, Zaber Ibn Abdul Hakim, Chris Thomas
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 the contagious loss (ablation) | LAMP · Mantis-Eval | ASR (%): 70.32 -> 73.43 | Improved |
| add the contagious loss (ablation) | LAMP · NLVR2 | ASR (%): 49.33 -> 52.73 | Improved |
| add the contagious loss (ablation) | LAMP · BLINK | ASR (%): 82.64 -> 84.86 | Improved |
| add the contagious loss and index-attention suppression to LM+dec+h losses (ablation) | LAMP · NLVR2 | ASR (%): 44.35 -> 52.73 | Improved |
| add contagious loss to the first three loss terms (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%): 68.66 -> 71.88 | Improved |
| add the hidden-state loss to the LM loss (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%): 59.55 -> 63.68 | Improved |
| add the Hausdorff attention loss to LM + hidden-state losses (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%): 63.68 -> 68.66 | Improved |
| increase number of universal perturbations from 1 to 2 (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%) | Improved |
| use more than 2 perturbations (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%) | No clear effect |
| increase perturbation budget beyond 12/255 (ablation) | LAMP · multi-image benchmarks | ASR (%) | No clear effect |
| set contagious and position losses weights to 1.2 instead of all 1 (ablation) | LAMP · Mantis-Eval, Mantis-CLIP | ASR (%): 72.9 -> 73.43 | No clear effect |
| add the contagious loss (contamination analysis) (ablation) | LAMP · Mantis-Eval | ASR (%): 61.1 -> 73.4 | Improved |
Closest papers by meaning
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Anyattack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models (CVPR 2025 · similarity 0.87)
- Attention! Your Vision Language Model Could Be Maliciously Manipulated (NeurIPS 2025 · similarity 0.87)
- Stage-wise Attention-Guided Region Sequencing for Adversarial Attacks on Large Vision-Language Models (NeurIPS 2026 · similarity 0.87)
- Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models (CVPR 2025 · similarity 0.87)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.86)
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.
