Adaptive Smoothness-weighted Adversarial Training for Multiple Perturbations with Its Stability Analysis

Jiancong Xiao, Zeyu Qin, Yanbo Fan, Baoyuan Wu, Jue Wang, Zhi-Quan Luo

arXiv: https://arxiv.org/abs/2210.00557

Workshop version: https://openreview.net/pdf?id=qvALKz8BUV

The Second Workshop on New Frontiers in Adversarial Machine Learning

Training Code

• train.py - Train the Adversarially Robust Models

  gpu_id - Id of GPU to be used - default = 0
model - Type of Adversarial Training: - default = 9
batch_size - Batch Size for Train Set -default = 128

• Adaptive Smoothness-weighted Adversarial Training

  python train.py -model 9

• The code is adopted from Locus Lab.

Citation

@article{xiao2022adaptive,
  title={Adaptive Smoothness-weighted Adversarial Training for Multiple Perturbations with Its Stability Analysis},
  author={Xiao, Jiancong and Qin, Zeyu and Fan, Yanbo and Wu, Baoyuan and Wang, Jue and Luo, Zhi-Quan},
  journal={arXiv preprint arXiv:2210.00557},
  year={2022}
}