Self-Paced-Adversarial-Training

Code for AAAI-23 paper "Improving Adversarial Robustness with Self-Paced Hard-Class Pair Reweighting"

Usage:

python train_cifar10.py

Requirements:

• python 3.7
• Pytorch 1.10
• Part of the code adopted from: TRADES (https://github.com/yaodongyu/TRADES)

Checkpoints:

• https://drive.google.com/file/d/1Nu7Ua-bhpVImoRhz7sV1mmtbpCJWYmBn/view?usp=sharing
• https://drive.google.com/file/d/18Swi_fqoybOK_Qabs7wbcdz4qCzlsDHP/view?usp=sharing

Reference:

@article{hou2022improving,
  title={Improving Adversarial Robustness with Self-Paced Hard-Class Pair Reweighting},
  author={Hou, Pengyue and Han, Jie and Li, Xingyu},
  journal={arXiv preprint arXiv:2210.15068},
  year={2022}
}