Defending against Backdoor Attack on Deep Neural Networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cheng, Hao, Xu, Kaidi, Liu, Sijia, Chen, Pin-Yu, Zhao, Pu, Lin, Xue
Format: Preprint
Published: 2020
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909552427925504
author Cheng, Hao
Xu, Kaidi
Liu, Sijia
Chen, Pin-Yu
Zhao, Pu
Lin, Xue
author_facet Cheng, Hao
Xu, Kaidi
Liu, Sijia
Chen, Pin-Yu
Zhao, Pu
Lin, Xue
contents Although deep neural networks (DNNs) have achieved a great success in various computer vision tasks, it is recently found that they are vulnerable to adversarial attacks. In this paper, we focus on the so-called \textit{backdoor attack}, which injects a backdoor trigger to a small portion of training data (also known as data poisoning) such that the trained DNN induces misclassification while facing examples with this trigger. To be specific, we carefully study the effect of both real and synthetic backdoor attacks on the internal response of vanilla and backdoored DNNs through the lens of Gard-CAM. Moreover, we show that the backdoor attack induces a significant bias in neuron activation in terms of the $\ell_\infty$ norm of an activation map compared to its $\ell_1$ and $\ell_2$ norm. Spurred by our results, we propose the \textit{$\ell_\infty$-based neuron pruning} to remove the backdoor from the backdoored DNN. Experiments show that our method could effectively decrease the attack success rate, and also hold a high classification accuracy for clean images.
format Preprint
id arxiv_https___arxiv_org_abs_2002_12162
institution arXiv
publishDate 2020
record_format arxiv
spellingShingle Defending against Backdoor Attack on Deep Neural Networks
Cheng, Hao
Xu, Kaidi
Liu, Sijia
Chen, Pin-Yu
Zhao, Pu
Lin, Xue
Cryptography and Security
Machine Learning
Although deep neural networks (DNNs) have achieved a great success in various computer vision tasks, it is recently found that they are vulnerable to adversarial attacks. In this paper, we focus on the so-called \textit{backdoor attack}, which injects a backdoor trigger to a small portion of training data (also known as data poisoning) such that the trained DNN induces misclassification while facing examples with this trigger. To be specific, we carefully study the effect of both real and synthetic backdoor attacks on the internal response of vanilla and backdoored DNNs through the lens of Gard-CAM. Moreover, we show that the backdoor attack induces a significant bias in neuron activation in terms of the $\ell_\infty$ norm of an activation map compared to its $\ell_1$ and $\ell_2$ norm. Spurred by our results, we propose the \textit{$\ell_\infty$-based neuron pruning} to remove the backdoor from the backdoored DNN. Experiments show that our method could effectively decrease the attack success rate, and also hold a high classification accuracy for clean images.
title Defending against Backdoor Attack on Deep Neural Networks
topic Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2002.12162