Title

GangSweep: Sweep Out Neural Backdoors by GAN

Document Type

Conference Paper

Publication Date

2020

DOI

10.1145/3394171.3413546

Publication Title

Proceedings of the 28th ACM International Conference on Multimedia

Pages

3173–3181

Conference Name

MM '20: The 28th ACM International Conference on Multimedia

Abstract

This work proposes GangSweep, a new backdoor detection framework that leverages the super reconstructive power of Generative Adversarial Networks (GAN) to detect and ''sweep out'' neural backdoors. It is motivated by a series of intriguing empirical investigations, revealing that the perturbation masks generated by GAN are persistent and exhibit interesting statistical properties with low shifting variance and large shifting distance in feature space. Compared with the previous solutions, the proposed approach eliminates the reliance on the access to training data, and shows a high degree of robustness and efficiency for detecting and mitigating a wide range of backdoored models with various settings. Moreover, this is the first work that successfully leverages generative networks to defend against advanced neural backdoors with multiple triggers and their polymorphic forms.

Comments

© Association for Computing Machinery

Open Access in ACM Digital Library.


Original Publication Citation

Zhu, L., Ning, R., Wang, C., Xin, C., & Wu, H. (2020). GangSweep: Sweep out neural backdoors by GAN. Proceedings of the 28th ACM International Conference on Multimedia, Seattle, Washington, USA. https://doi.org/10.1145/3394171.3413546.

Share

COinS