tree: f893a7e181cc61b17ce942d7492a1f63911047ab [path history] [tgz]
  1. README.md
  2. VAE.py
  3. VAE_example.ipynb
example/autoencoder/variational_autoencoder/README.md

Variational Auto Encoder(VAE)

This folder contains a tutorial which implements the Variational Auto Encoder in MXNet using the MNIST handwritten digit recognition dataset. Model built is referred from Auto-Encoding Variational Bayes paper. This paper introduces a stochastic variational inference and learning algorithm that scales to large datasets.

Prerequisites: To run this example, you need:

Files in this folder:

  • VAE_example.ipynb : Jupyter notebook which explains concept of VAE step by step and also shows how to use MXNet-based VAE class(from VAE.py) to do the training directly.

  • VAE.py : Contains class which implements the Variational Auto Encoder. This is used in the above tutorial.

In VAE, the encoder becomes a variational inference network that maps the data to a distribution for the hidden variables, and the decoder becomes a generative network that maps the latent variables back to the data. The network architecture shown in the tutorial uses Gaussian MLP as an encoder and Bernoulli MLP as a decoder.