Awesome

Dynamic AdveRsarial Training (Dynamic/DART)

Code for ICML2019 Paper "On the Convergence and Robustness of Adversarial Training"

One Important Message in this paper: To ensure better robustness, it is essential to use adversarial examples with better convergence quality at the later stages of training. Yet at the early stages, high convergence quality adversarial examples are not necessary and may even lead to poor robustness.

Convergence quality is measured by First-Order Stationary Condition (FOSC) <img src="https://github.com/YisenWang/dynamic_adv_training/blob/master/fosc.png" width="50%" height="50%">

Requirements

• Python 3.5.2,
• Tensorflow 1.10.1
• Keras 2.2.2

Usage

Simply run the code by: python3 train_models.py

Citing this work

If you use this code in your work, please cite the accompanying paper:

@inproceedings{wang2019dynamic,
  title={On the Convergence and Robustness of Adversarial Training},
  author={Wang, Yisen and Ma, Xingjun and Bailey, James and Yi, Jinfeng and Zhou, Bowen and Gu, Quanquan},
  booktitle={International Conference on Machine Learning},
  year={2019}
}