This repository contains the code for the following paper
Exploring Memorization in Adversarial Training (ICLR 2022)
Yinpeng Dong, Ke Xu, Xiao Yang, Tianyu Pang, Zhijie Deng, Hang Su, and Jun Zhu
If you find our methods useful, please consider citing:
@inproceedings{dong2022exploring,
title={Exploring Memorization in Adversarial Training},
author={Yinpeng Dong and Ke Xu and Xiao Yang and Tianyu Pang and Zhijie Deng and Hang Su and Jun Zhu},
booktitle={International Conference on Learning Representations},
year={2022}
}
- Python (3.6.8)
- Pytorch (1.3.0)
- torchvision (0.4.1)
- numpy
For PGD-AT
python train.py --wd 0 --noise-type label_symmetric --noise-rate 1.0
For TRAEDS
python train.py --wd 0 --noise-type label_symmetric --noise-rate 1.0 --loss-type trades
For PGD-AT + TE
python train_te.py
For TRADES + TE
python train_te.py --loss-type trades