Adversarial training for free!

Authors: Ali Shafahi, Mahyar Najibi, Mohammad Amin Ghiasi, Zheng Xu, John Dickerson, Christoph Studer, Larry S. Davis, Gavin Taylor, Tom Goldstein

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our free adversarial training algorithm achieves comparable robustness to PGD adversarial training on the CIFAR-10 and CIFAR-100 datasets at negligible additional cost compared to natural training, and can be 7 to 30 times faster than other strong adversarial training methods. Using a single workstation with 4 P100 GPUs and 2 days of runtime, we can train a robust model for the large-scale Image Net classification task that maintains 40% accuracy against PGD attacks.
Researcher Affiliation Academia Ali Shafahi University of Maryland ashafahi@cs.umd.edu Mahyar Najibi University of Maryland najibi@cs.umd.edu Amin Ghiasi University of Maryland amin@cs.umd.edu Zheng Xu University of Maryland xuzh@cs.umd.edu John Dickerson University of Maryland john@cs.umd.edu Christoph Studer Cornell University studer@cornell.edu Larry S. Davis University of Maryland lsd@umiacs.umd.edu Gavin Taylor United States Naval Academy taylor@usna.edu Tom Goldstein University of Maryland tomg@cs.umd.edu
Pseudocode Yes Algorithm 1 Free Adversarial Training (Free-m)
Open Source Code Yes Adversarial Training for Free code for CIFAR-10 in TensorFlow can be found here: https://github. com/ashafahi/free_adv_train/ Image Net Adversarial Training for Free code in Pytorch can be found here: https://github.com/ mahyarnajibi/Free Adversarial Training
Open Datasets Yes Our free adversarial training algorithm achieves comparable robustness to PGD adversarial training on the CIFAR-10 and CIFAR-100 datasets at negligible additional cost compared to natural training... Image Net is a large image classification dataset of over 1 million high-res images and 1000 classes (Russakovsky et al. [2015]).
Dataset Splits Yes We train various CIFAR-10 models using the Wide-Resnet 32-10 model and standard hyperparameters used by Madry et al. [2017]. ...CIFAR-10 and CIFAR-100 models that are 7-PGD adversarially trained have natural accuracies of 87.25% and 59.87%, respectively. ...Image Net is a large image classification dataset of over 1 million high-res images and 1000 classes (Russakovsky et al. [2015]).
Hardware Specification Yes Using a single workstation with 4 P100 GPUs and 2 days of runtime, we can train a robust model for the large-scale Image Net classification task... Free training on Res Net-101 and Res Net-152 each take roughly 1.7 and 2.4 more time than Res Net-50 on the same machine, respectively.
Software Dependencies No The paper mentions the use of 'TensorFlow' for CIFAR-10 code and 'Pytorch' for ImageNet code in footnotes. However, it does not specify any version numbers for these frameworks or any other software dependencies, which are required for reproducibility.
Experiment Setup Yes In the proposed method (alg. 1), we repeat (i.e. replay) each minibatch m times before switching to the next minibatch. ...In all experiments, the training batch size was 256. ...To craft attacks, we used a step-size of 1 and the corresponding ϵ used during training.