DAN: Distributional Adversarial Networks

Tensorflow demo code for paper Distributional Adversarial Networks by Chengtao Li*, David Alvarez-Melis*, Keyulu Xu, Stefanie Jegelka and Suvrit Sra.

Summary

The main difference with the original GAN method is that the Discriminator is operates on samples (of n>1 examples) rather than a single sample point to discriminate between real and generated distributions. In the paper we propose two such type of methods:

• A single-sample classifier $M_S$ which classifies samples as fake or real (i.e. a sample-based analogue to the original GAN classifier)
• A two-sample discriminator $M_{2S}$ which must decide whether two samples are drawn from the same distribution or not (reminiscent of two-sample tests in the the statistics literature)

Both of these methods relies on a first stage encoder (Deep Mean Encoder), which embeds and aggregates individual examples to obtain a fixed-size representation of the sample. These vectors are then used as inputs to the two types of classifiers.

A schematic representation of these two methods is:

Prerequisites

• Python 2.7
• tensorflow >= 1.0
• numpy
• scipy
• matplotlib

Toy Experiments

A self-contained implementation of the two DAN models applied to a simple 2D mixture of gaussians examples can be found in this notebook in toy folder. Some of the visualization tools were borrowed from here.

Visualization

Vanilla GAN	DAN-S

DAN-2S	Ground Truth

MNIST Digit Generation

This part of code can be used to reproduce experimental results on MNIST digit generation. It lies in mnist folder and is built based on DCGAN Implementation.

Training

To train the adversarial network, run

python main_mnist.py --model_mode [MODEL_MODE] --is_train True

Here MODEL_MODE can be one of gan (for vanilla GAN model), dan_s (for DAN-S) or dan_2s (for DAN-2S).

Evaluation

To evaluate how well the model recovers the mode frequencies, one need an accurate classifier on MNIST dataset as an approximate label indicator. The code for the classifier is in mnist_classifier.py and is adapted from Tensorflow-Examples. To train the classifier, run

python mnist_classifier.py

The classifier has an accuracy of ~97.6% on test set after 10 epochs and is stored in the folder mnist_cnn for later evaluation. To use the classifier to estimate the label frequencies of generated figures, run

python main_mnist.py --model_mode [MODEL_MODE] --is_train False

The result will be saved to the file specified by savepath. A random run gives the following results with different model_mode's.

Visualization

The following visualization shows how the randomly generated figures evolve through 100 epochs with different models. While for vanilla GAN the figures mostly concentrate on ''easy-to-generate'' modes like 1, models within DAN framework generate figures that have better coverages over different modes.

Vanilla GAN	DAN-S	DAN-2S

Domain Adaptation

This part of code can be used to reproduce experimental results of domain adaptation from MNIST to MNIST-M. It lies in dann folder and is built based on DANN Implementation.

Build Dataset

Run the following commands to download and create MNIST-M dataset.

curl -O http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz
python create_mnistm.py

(instructions from here)

Training

To train the adversarial network, run

python mnist_dann.py --model_mode [MODEL_MODE]

Here MODEL_MODE can be one of gan (for vanilla GAN model), dan_s (for DAN-S) or dan_2s (for DAN-2S). A random run with different different modes gives the following prediction accuracy on MNIST-M when the classifier is trained on MNIST.

Citation

If you use this code for your research, please cite our paper:

@article{li2017distributional,
  title={Distributional Adversarial Networks},
  author={Chengtao Li, David Alvarez-Melis, Keyulu Xu, Stefanie Jegelka, Suvrit Sra},
  journal={arXiv preprint arXiv:1706.09549},
  year={2017}
}

Contact

Please email to [email protected] should you have any questions, comments or suggestions.