DiffSmooth: Certifiably Robust Learning via Diffusion Models and Local Smoothing
Published in 32nd USENIX Security Symposium (USENIX Security 23), 2023
DiffSmooth combines the power of diffusion models with local smoothing techniques to achieve certifiably robust learning. This work provides both theoretical guarantees and practical defense mechanisms against adversarial attacks.
*Equal contribution
Recommended citation: Jiawei Zhang*, Zhongzhu Chen*, Huan Zhang, Chaowei Xiao, and Bo Li. (2023). "DiffSmooth: Certifiably Robust Learning via Diffusion Models and Local Smoothing." USENIX Security 23, pp. 4787-4804.
Download Paper