Population Matching Discrepancy

Code for the paper Population Matching Discrepancy and Applications in Deep Learning.

Installation

Make sure that you have valid c++11 and cuda compilers.

git clone git@github.com:cjf00000/pmd.git
cd pmd
pip install --upgrade pip
pip install cython numpy tensorflow scipy matplotlib scikit-image scikit-learn seaborn
pip install -e .

Domain Adaptation Usage

Download the data

cd domain_adaptation
wget -O data.tar.gz https://github.com/cjf00000/pmd-data/blob/master/data.tar.gz?raw=true
tar xzvf data.tar.gz

See config.pmd or config.mmd for the training recipes.

Generative Model Usage

The scripts will automatically download MNIST and CIFAR10 datasets.

Running:

cd generative-model
mkdir data
configs/mnist_pmd_fc

The images will be generated in the result directory.

Cite

If you find the code is useful, please cite our paper!

@inproceedings{chen2017population,
  title={Population Matching Discrepancy and Applications in Deep Learning},
  author={Chen, Jianfei and Chongxuan, LI and Ru, Yizhong and Zhu, Jun},
  booktitle={Advances in Neural Information Processing Systems},
  pages={6263--6275},
  year={2017}
}