Generative Moment Matching Networks(GMMN)

---

Introduction

• The code is widely adapted from two nice repositories:
  https://github.com/Abhipanda4/GMMN-Pytorch
  https://github.com/siddharth-agrawal/Generative-Moment-Matching-Networks.
  
• I failed to implement the greedy layer-wise pretraining and fine-tuning scheme mentioned in the paper as I have not fully understood it.

Environment & Requirements

• CentOS Linux release 7.2.1511 (Core)
  
• python 3.6.5
  
• pytorch 1.0.0
  
• torchvision
  
• argparse
  
• pickle

Architecture

Usage

1. Train the networks using MNIST:

python train.py

Two folders will be created, i.e., ./data & ./models. As their names imply, they are used to store data and trained models, repectively.

2. Visualize the outputs of the gmmn:

python Visualize.py --visualize gmmn

3. Visualize the outputs of the autoencoder:

python Visualize.py --visualize autoencoder

References

https://github.com/Abhipanda4/GMMN-Pytorch
https://github.com/siddharth-agrawal/Generative-Moment-Matching-Networks