Feature Denoising for Improving Adversarial Robustness

Adversarial attacks to image classification systems present challenges to convolutional networks and opportunities for understanding them. This study suggests that adversarial perturbations on images lead to noise in the features constructed by these networks. Motivated by this observation, we devel...

Full description

Saved in:
Bibliographic Details
Published in:Proceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) pp. 501 - 509
Main Authors: Xie, Cihang, Wu, Yuxin, Maaten, Laurens van der, Yuille, Alan L., He, Kaiming
Format: Conference Proceeding
Language:English
Published: IEEE 01.06.2019
Subjects:
Accuracy
Categorization
Computational modeling
Computer architecture
Convolutional neural networks
Deep Learning
Glass box
Noise reduction
Perturbation methods
Recognition: Detection
Retrieval
Robustness
Testing
Training
ISSN:1063-6919
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Adversarial attacks to image classification systems present challenges to convolutional networks and opportunities for understanding them. This study suggests that adversarial perturbations on images lead to noise in the features constructed by these networks. Motivated by this observation, we develop new network architectures that increase adversarial robustness by performing feature denoising. Specifically, our networks contain blocks that denoise the features using non-local means or other filters; the entire networks are trained end-to-end. When combined with adversarial training, our feature denoising networks substantially improve the state-of-the-art in adversarial robustness in both white-box and black-box attack settings. On ImageNet, under 10-iteration PGD white-box attacks where prior art has 27.9% accuracy, our method achieves 55.7%; even under extreme 2000-iteration PGD white-box attacks, our method secures 42.6% accuracy. Our method was ranked first in Competition on Adversarial Attacks and Defenses (CAAD) 2018 --- it achieved 50.6% classification accuracy on a secret, ImageNet-like test dataset against 48 unknown attackers, surpassing the runner-up approach by ~10%. Code is available at https://github.com/facebookresearch/ImageNet-Adversarial-Training.
ISSN:1063-6919
DOI:10.1109/CVPR.2019.00059