Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoff
Authors:
- 6 June 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 3855-3859
Abstract
Data poisoning and backdoor attacks manipulate victim models by maliciously modifying training data. In light of this growing threat, a recent survey of industry professionals revealed heightened fear in the private sector regarding data poisoning. Many previous defenses against poisoning either fail in the face of increasingly strong attacks, or they significantly degrade performance. However, we find that strong data augmentations, such as mixup and CutMix, can significantly diminish the threat of poisoning and backdoor attacks without trading off performance. We further verify the effectiveness of this simple defense against adaptive poisoning methods, and we compare to baselines including the popular differentially private SGD (DP-SGD) defense. In the context of backdoors, CutMix greatly mitigates the attack while simultaneously increasing validation accuracy by 9%.Keywords
All Related Versions
This publication has 7 references indexed in Scilit:
- Witchcraft: Efficient PGD Attacks with Random Step Size2020
- Adversarial Machine Learning - Industry PerspectivesSSRN Electronic Journal, 2020
- CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features2019
- DeepISP: Toward Learning an End-to-End Image Processing PipelineIEEE Transactions on Image Processing, 2018
- A Marauder's Map of Security and Privacy in Machine Learning2018
- Deep Learning with Differential Privacy2016
- ANTIDOTE2009