GMVAE

This repository contains a TensorFlow implementation of an unsupervised Gaussian Mixture Variational Autoencoder (GMVAE) on the MNIST dataset, specifically making use of the Probability library.

There are currently three models in use:

• VAE is a standard implementation of the Variational Autoencoder, with no convolutional layers
• VAE_GMP is an adaptation of VAE to make use of a Gaussian Mixture prior, instead of a standard Normal distribution
• GMVAE is an attempt to replicate the work described in this blog and inspired from this paper

The directory layout is as follows:

• bin: Bash example scripts for running the aforementioned models
• checkpoints: Directories to save checkpoints of trained model states
• scripts: TensorFlow scripts to implement the models and run them using the main run_gmvae.py script, alongside other helpful modules (helpers.py and base.py)

Note: This is a work in progress, so any contributions/feedback will be well-received.

Dependencies

• TensorFlow 1.13.1
• TensorFlow Datasets
• TensorFlow Probability 0.6.0
• Cuda 10.0
• Cudnn 7.4.2