Adversarial Dropout Regularization

This is the implementation of Adversarial Dropout Regularization in Pytorch. The code is written by Kuniaki Saito. The work was accepted by ICLR 2018.

Adversarial Dropout Reguralization: [Project] [Paper].

Getting Started

Installation

• Install PyTorch (Works on Version 0.2.0_3) and dependencies from http://pytorch.org.
• Due to the change of calculation of kl divergence, it may not work for newer version.
• Install Torch vision from the source.
• Install torchnet as follows

pip install git+https://github.com/pytorch/tnt.git@master

Download Dataset

Download MNIST Dataset here. Resized image dataset is contained in the file. Place it in the directory ./data.

ADR train/evaluation

For example, adaptation from svhn to mnist.

python main.py --source svhn --target mnist

In case of version conflict of Pytorch, use the option --use_abs_diff, which will change the measurement from kl divergence to absolute difference.

python main.py --source svhn --target mnist --use_abs_diff

Citation

If you use this code for your research, please cite our papers.

@article{saito2017adversarial,
  title={Adversarial Dropout Regularization},
  author={Saito, Kuniaki and Ushiku, Yoshitaka and Harada, Tatsuya and Saenko, Kate},
  journal={arXiv preprint arXiv:1711.01575},
  year={2017}
}

License

MIT