Learning Sparse Latent Representations with the Deep Copula IB

This repository includes a basic implementation of our paper Learning Sparse Latent Representations with the Deep Copula IB.

Installation

The code was tested under Python 2.7.12. To run the code, it is necessary to install the following dependencies from the requirements file.

Open a new terminal and create a virtualenv:

mkdir copulaib
cd copulaib
git clone https://github.com/wieserm/copula-ib-public.git
cd ..
virtualenv copulaib/paper

Activate the environment:

source copulaib/paper/bin/activate

Install the dependencies:

pip install -r copulaib/copula-ib-public/requirements.txt

Run the code

To run the code for the artificial experiment please execute the following command:

python Main.py

The result is an information curve plot with the used latent dimensions.

Reference

The paper "Learning Sparse Latent Representations with the Deep Copula IB" has been accepted to ICLR 2018. If you like it please cite us.

@ARTICLE{WieczorekWieser,
   author = {{Wieczorek}, A. and {Wieser}, M. and {Murezzan}, D. and {Roth}, V.},
   title = "{Learning Sparse Latent Representations with the Deep Copula Information Bottleneck}",
   journal = {ArXiv e-prints},
   eprint = {1804.06216},
   year = 2018,
   month = apr,
}