Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift

Diffusion models (DM) have become state-of-the-art generative models because of their capability of generating high-quality images from noises without adversarial training. However, they are vulnerable to backdoor attacks as reported by recent studies. When a data input (e.g., some Gaussian noise) is stamped with a trigger (e.g., a white patch), the backdoored model always generates the target image (e.g., an improper photo). However, effective defense strategies to mitigate backdoors from DMs are underexplored. To bridge this gap, we propose the first backdoor detection and removal framework for DMs. We evaluate our framework Elijah on over hundreds of DMs of 3 types including DDPM, NCSN and LDM, with 13 samplers against 3 existing backdoor attacks. Extensive experiments show that our approach can have close to 100% detection accuracy and reduce the backdoor effects to close to zero without significantly sacrificing the model utility.

Authors

• Shengwei An Purdue University
• Sheng-Yen Chou The Chinese University of Hong Kong
• Kaiyuan Zhang Purdue University
• Qiuling Xu Purdue University
• Guanhong Tao Purdue University
• Guangyu Shen Purdue University
• Siyuan Cheng Zhejiang University Zhejiang University - Ant Group Joint Laboratory of Knowledge Graph Platform and Content Group, Tencent
• Shiqing Ma University of Massachusetts Amherst
• Pin-Yu Chen IBM Research
• Tsung-Yi Ho Department of Computer Science and Engineering, The Chinese University of Hong Kong
• Xiangyu Zhang Megvii Technology Beijing Academy of Artifcial Intelligence

Keywords

• CV: Adversarial Attacks & Robustness
• ML: Adversarial Learning & Robustness
• ML: Deep Generative Models & Autoencoders
• PEAI: Safety, Robustness & Trustworthiness