Guardado en:
| Autores principales: | , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2412.16512 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909437728391168 |
|---|---|
| author | Qi, Zhengyang Xu, Xiaohua |
| author_facet | Qi, Zhengyang Xu, Xiaohua |
| contents | Flow-based generative models (FMs) have rapidly advanced as a method for mapping noise to data, its efficient training and sampling process makes it widely applicable in various fields. FMs can be viewed as a variant of diffusion models (DMs). At the same time, previous studies have shown that DMs are vulnerable to Trojan/Backdoor attacks, a type of output manipulation attack triggered by a maliciously embedded pattern at model input. We found that Trojan attacks on generative models are essentially equivalent to image transfer tasks from the backdoor distribution to the target distribution, the unique ability of FMs to fit any two arbitrary distributions significantly simplifies the training and sampling setups for attacking FMs, making them inherently natural targets for backdoor attacks. In this paper, we propose TrojFlow, exploring the vulnerabilities of FMs through Trojan attacks. In particular, we consider various attack settings and their combinations and thoroughly explore whether existing defense methods for DMs can effectively defend against our proposed attack scenarios. We evaluate TrojFlow on CIFAR-10 and CelebA datasets, our experiments show that our method can compromise FMs with high utility and specificity, and can easily break through existing defense mechanisms. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_16512 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | TrojFlow: Flow Models are Natural Targets for Trojan Attacks Qi, Zhengyang Xu, Xiaohua Computer Vision and Pattern Recognition Artificial Intelligence Flow-based generative models (FMs) have rapidly advanced as a method for mapping noise to data, its efficient training and sampling process makes it widely applicable in various fields. FMs can be viewed as a variant of diffusion models (DMs). At the same time, previous studies have shown that DMs are vulnerable to Trojan/Backdoor attacks, a type of output manipulation attack triggered by a maliciously embedded pattern at model input. We found that Trojan attacks on generative models are essentially equivalent to image transfer tasks from the backdoor distribution to the target distribution, the unique ability of FMs to fit any two arbitrary distributions significantly simplifies the training and sampling setups for attacking FMs, making them inherently natural targets for backdoor attacks. In this paper, we propose TrojFlow, exploring the vulnerabilities of FMs through Trojan attacks. In particular, we consider various attack settings and their combinations and thoroughly explore whether existing defense methods for DMs can effectively defend against our proposed attack scenarios. We evaluate TrojFlow on CIFAR-10 and CelebA datasets, our experiments show that our method can compromise FMs with high utility and specificity, and can easily break through existing defense mechanisms. |
| title | TrojFlow: Flow Models are Natural Targets for Trojan Attacks |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2412.16512 |