Code for NeurIPS 2021 Paper "Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks" by Hanxun Huang, Yisen Wang, Sarah Monazam Erfani, Quanquan Gu, James Bailey, Xingjun Ma
- Model defined in models/RobustWideResNet.py
def RobustWideResNet34(num_classes=10):
# WRN-34-R configurations
return RobustWideResNet(
num_classes=num_classes, channel_configs=[16, 320, 640, 512],
depth_configs=[5, 5, 5], stride_config=[1, 2, 2], stem_stride=1,
drop_rate_config=[0.0, 0.0, 0.0], zero_init_residual=False,
block_types=['basic_block', 'basic_block', 'basic_block'],
activations=['ReLU', 'ReLU', 'ReLU'], is_imagenet=False,
use_init=True)- Pretrained Weights for WRN-34-R used in Table 2 available on Google Drive
- All hyperparameters/settings for each model/method used in Table 2 are stored in configs/*.yaml files.
Replace PGD with other attacks ['CW', 'GAMA', 'AA'].
python main.py --config_path configs/config-WRN-34-R
--exp_name /path/to/experiments/folders
--version WRN-34-R-trades
--load_best_model --attack PGD --data_parallelReplace PGD with other attacks ['CW', 'GAMA', 'AA'].
python main.py --config_path configs/config-WRN-34-R
--exp_name /path/to/experiments/folders
--version WRN-34-R-trades-500k
--load_best_model --attack PGD --data_parallelpython main.py --config_path configs/config-WRN-34-R
--exp_name /path/to/experiments/folders
--version WRN-34-R-trades-500k
--train --data_parallel- Note: This is not maintained, please find up-to-date leaderboard is available in RobustBench.
| # | paper | model | architecture | clean | report. | AA |
|---|---|---|---|---|---|---|
| 1 | (Gowal et al., 2020)‡ | available | WRN-70-16 | 91.10 | 65.87 | 65.88 |
| 2 | Ours‡ + EMA | available | WRN-34-R | 91.23 | 62.54 | 62.54 |
| 3 | Ours‡ | available | WRN-34-R | 90.56 | 61.56 | 61.56 |
| 4 | (Wu et al., 2020a)‡ | available | WRN-34-15 | 87.67 | 60.65 | 60.65 |
| 5 | (Wu et al., 2020b)‡ | available | WRN-28-10 | 88.25 | 60.04 | 60.04 |
| 6 | (Carmon et al., 2019)‡ | available | WRN-28-10 | 89.69 | 62.5 | 59.53 |
| 7 | (Sehwag et al., 2020)‡ | available | WRN-28-10 | 88.98 | - | 57.14 |
| 8 | (Wang et al., 2020)‡ | available | WRN-28-10 | 87.50 | 65.04 | 56.29 |
@inproceedings{huang2021exploring,
title={Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks},
author={Hanxun Huang and Yisen Wang and Sarah Monazam Erfani and Quanquan Gu and James Bailey and Xingjun Ma},
booktitle={NeurIPS},
year={2021}
}