LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915760602873856 |
|---|---|
| author | Ishmam, Alvi Md Sarker, Najibul Haque Hakim, Zaber Ibn Abdul Thomas, Chris |
| author_facet | Ishmam, Alvi Md Sarker, Najibul Haque Hakim, Zaber Ibn Abdul Thomas, Chris |
| contents | Multimodal Large Language Models (MLLMs) have achieved remarkable performance across vision-language tasks. Recent advancements allow these models to process multiple images as inputs. However, the vulnerabilities of multi-image MLLMs remain unexplored. Existing adversarial attacks focus on single-image settings and often assume a white-box threat model, which is impractical in many real-world scenarios. This paper introduces LAMP, a black-box method for learning Universal Adversarial Perturbations (UAPs) targeting multi-image MLLMs. LAMP applies an attention-based constraint that prevents the model from effectively aggregating information across images. LAMP also introduces a novel cross-image contagious constraint that forces perturbed tokens to influence clean tokens, spreading adversarial effects without requiring all inputs to be modified. Additionally, an index-attention suppression loss enables a robust position-invariant attack. Experimental results show that LAMP outperforms SOTA baselines and achieves the highest attack success rates across multiple vision-language tasks and models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_21220 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models Ishmam, Alvi Md Sarker, Najibul Haque Hakim, Zaber Ibn Abdul Thomas, Chris Computer Vision and Pattern Recognition Multimodal Large Language Models (MLLMs) have achieved remarkable performance across vision-language tasks. Recent advancements allow these models to process multiple images as inputs. However, the vulnerabilities of multi-image MLLMs remain unexplored. Existing adversarial attacks focus on single-image settings and often assume a white-box threat model, which is impractical in many real-world scenarios. This paper introduces LAMP, a black-box method for learning Universal Adversarial Perturbations (UAPs) targeting multi-image MLLMs. LAMP applies an attention-based constraint that prevents the model from effectively aggregating information across images. LAMP also introduces a novel cross-image contagious constraint that forces perturbed tokens to influence clean tokens, spreading adversarial effects without requiring all inputs to be modified. Additionally, an index-attention suppression loss enables a robust position-invariant attack. Experimental results show that LAMP outperforms SOTA baselines and achieves the highest attack success rates across multiple vision-language tasks and models. |
| title | LAMP: Learning Universal Adversarial Perturbations for Multi-Image Tasks via Pre-trained Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.21220 |