Test this code online with Binder.

itl-ae

Information Theoretic Auto-Encoder, Eder Santana, Matthew Emigh, Jose C Principe. IJCNN 2016 (accepted)

Installation

1. Install an Anaconda environment (recommended).
2. From the environment, update Theano: conda uninstall theano; pip install https://github.com/Theano/Theano.git
3. Install Keras: pip install https://github.com/fchollet/keras.git@52f48e1f462090db19b03ae11dce
4. Install Seya: pip install https://github.com/EderSantana/seya.git@5ce4211648a24ee5514df70a
5. Have fun with the Jupyter notebooks!

Notebooks

ITL AE best log-likelihood and ITL AE-Swiss Roll were the ones used to generate the results on the paper, the other ones show work in progress not published yet that you may find interesting.

Torch version

For torch users, there is an implementation here. It does not reproduce all the results in the paper like this one.