Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909372092776448 |
|---|---|
| author | Huang, Youcheng Zhu, Fengbin Tang, Jingkun Zhou, Pan Lei, Wenqiang Lv, Jiancheng Chua, Tat-Seng |
| author_facet | Huang, Youcheng Zhu, Fengbin Tang, Jingkun Zhou, Pan Lei, Wenqiang Lv, Jiancheng Chua, Tat-Seng |
| contents | Visual Language Models (VLMs) are vulnerable to adversarial attacks, especially those from adversarial images, which is however under-explored in literature. To facilitate research on this critical safety problem, we first construct a new laRge-scale Adervsarial images dataset with Diverse hArmful Responses (RADAR), given that existing datasets are either small-scale or only contain limited types of harmful responses. With the new RADAR dataset, we further develop a novel and effective iN-time Embedding-based AdveRSarial Image DEtection (NEARSIDE) method, which exploits a single vector that distilled from the hidden states of VLMs, which we call the attacking direction, to achieve the detection of adversarial images against benign ones in the input. Extensive experiments with two victim VLMs, LLaVA and MiniGPT-4, well demonstrate the effectiveness, efficiency, and cross-model transferrability of our proposed method. Our code is available at https://github.com/mob-scu/RADAR-NEARSIDE |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_22888 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector Huang, Youcheng Zhu, Fengbin Tang, Jingkun Zhou, Pan Lei, Wenqiang Lv, Jiancheng Chua, Tat-Seng Computer Vision and Pattern Recognition Computation and Language Cryptography and Security Visual Language Models (VLMs) are vulnerable to adversarial attacks, especially those from adversarial images, which is however under-explored in literature. To facilitate research on this critical safety problem, we first construct a new laRge-scale Adervsarial images dataset with Diverse hArmful Responses (RADAR), given that existing datasets are either small-scale or only contain limited types of harmful responses. With the new RADAR dataset, we further develop a novel and effective iN-time Embedding-based AdveRSarial Image DEtection (NEARSIDE) method, which exploits a single vector that distilled from the hidden states of VLMs, which we call the attacking direction, to achieve the detection of adversarial images against benign ones in the input. Extensive experiments with two victim VLMs, LLaVA and MiniGPT-4, well demonstrate the effectiveness, efficiency, and cross-model transferrability of our proposed method. Our code is available at https://github.com/mob-scu/RADAR-NEARSIDE |
| title | Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector |
| topic | Computer Vision and Pattern Recognition Computation and Language Cryptography and Security |
| url | https://arxiv.org/abs/2410.22888 |