Skip to content

Latest commit

 

History

History
49 lines (44 loc) · 1.76 KB

README.md

File metadata and controls

49 lines (44 loc) · 1.76 KB
Raw

Efficient and Effective Augmentation Strategy for Adversarial Training

This repository contains codes for the training and evaluation of our NeurIPS-22 paper Efficient and Effective Strategy for Adversarial Training. The openreview link for the paper is also available.

plot

Environment Settings

  • Python 3.6.9
  • PyTorch 1.8
  • Torchvision 0.8.0
  • Numpy 1.19.2

The checkpoints can be found at Google Drive

Training

For training DAJAT:

python train_DAJAT.py --use_defaults ['NONE','CIFAR10_RN18', 'CIFAR10_WRN','CIFAR100_WRN', 'CIFAR100_RN18']

For training ACAT:

python train_DAJAT.py --use_defaults ['NONE','CIFAR10_RN18', 'CIFAR10_WRN','CIFAR100_WRN', 'CIFAR100_RN18']  --num_autos 0 --epochs 110 --beta --train_budget 'low'

Evaluation

The GAMA-PGD-100 evaluation code is provided in eval.py. For evaluation of the trained model:

python eval.py --trained_model 'PATH OF TRAINED MODEL'

Further all the running details are provided in run.sh. It is recommended to use this file for training and evaluation of DAJAT.

Results

plot plot

Results obtained using higher number of attack steps and 200 epochs for training:

Citing this work

@inproceedings{
addepalli2022efficient,
title={Efficient and Effective Augmentation Strategy for Adversarial Training},
author={Sravanti Addepalli and Samyak Jain and Venkatesh Babu Radhakrishnan},
booktitle={Advances in Neural Information Processing Systems},
editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
year={2022},
url={https://openreview.net/forum?id=ODkBI1d3phW}
}