English

Backdoor Defense via Suppressing Model Shortcuts

Computer Vision and Pattern Recognition 2023-03-07 v2 Cryptography and Security Machine Learning

Abstract

Recent studies have demonstrated that deep neural networks (DNNs) are vulnerable to backdoor attacks during the training process. Specifically, the adversaries intend to embed hidden backdoors in DNNs so that malicious model predictions can be activated through pre-defined trigger patterns. In this paper, we explore the backdoor mechanism from the angle of the model structure. We select the skip connection for discussions, inspired by the understanding that it helps the learning of model `shortcuts' where backdoor triggers are usually easier to be learned. Specifically, we demonstrate that the attack success rate (ASR) decreases significantly when reducing the outputs of some key skip connections. Based on this observation, we design a simple yet effective backdoor removal method by suppressing the skip connections in critical layers selected by our method. We also implement fine-tuning on these layers to recover high benign accuracy and to further reduce ASR. Extensive experiments on benchmark datasets verify the effectiveness of our method.

Keywords

Cite

@article{arxiv.2211.05631,
  title  = {Backdoor Defense via Suppressing Model Shortcuts},
  author = {Sheng Yang and Yiming Li and Yong Jiang and Shu-Tao Xia},
  journal= {arXiv preprint arXiv:2211.05631},
  year   = {2023}
}

Comments

This paper is accepted by ICASSP 2023. 5 pages

R2 v1 2026-06-28T05:36:23.517Z