NeurIPS'21: Probabilistic Margins for Instance Reweighting in Adversarial Training (Pytorch implementation).

========

This is the code for the paper:

Probabilistic Margins for Instance Reweighting in Adversarial Training

Qizhou Wang*, Feng Liu*, Bo Han, Tongliang Liu, Chen Gong, Gang Niu, Mingyuan Zhou, Masashi Sugiyama

To be presented at NeurIPS 2021.

If you find this code useful in your research then please cite

@inproceedings{wang2021probabilistic,
title={Probabilistic Margins for Instance Reweighting in Adversarial Training},
author={Qizhou Wang and Feng Liu and Bo Han and Tongliang Liu and Chen Gong and Gang Niu and Mingyuan Zhou and Masashi Sugiyama},
booktitle={NeurIPS},
year={2021}
}

All code was developed and tested on a single machine equiped with a NVIDIA GTX3090 GPU. The environment is as bellow:

• Ubuntu 18.04

• CUDA 10.2.89

• Python 3.7.6 (Anaconda 4.9.2 64 bit)

• PyTorch 1.5.0

• numpy 1.18.1

python train.py --method mail_at --bias -0.5 --slope 10

python train.py --method mail_trades --bias 0  --slope 2

Contact: Qizhou Wang ([email protected]); Feng Liu ([email protected]).