VAE with Volume-Preserving Flows

This is a PyTorch implementation of two volume-preserving flows as described in the following papers:

• Tomczak, J. M., & Welling, M., Improving Variational Auto-Encoders using Householder Flow, arXiv preprint, 2016
• Tomczak, J. M., & Welling, M., Improving Variational Auto-Encoders using convex combination linear Inverse Autoregressive Flow, arXiv preprint, 2017

Data

The experiments can be run on four datasets:

• static MNIST: links to the datasets can found at link;
• binary MNIST: the dataset is loaded from Keras;
• OMNIGLOT: the dataset could be downloaded from link;
• Caltech 101 Silhouettes: the dataset could be downloaded from link.

Run the experiment

1. Set-up your experiment in experiment.py.
2. Run experiment:

python experiment.py

Models

You can run a vanilla VAE, a VAE with the Householder Flow (HF) or the convex combination linear Inverse Autoregressive Flow (ccLinIAF) by setting model_name argument to either vae, vae_HF or vae_ccLinIAF, respectively. Setting number_combination for vae_ccLinIAF to 1 results in vae_linIAF.

Citation

Please cite our paper if you use this code in your research:

@article{TW:2017,
  title={{Improving Variational Auto-Encoders using convex combination linear Inverse Autoregressive Flow}},
  author={Tomczak, Jakub M and Welling, Max},
  journal={arXiv},
  year={2017}
}

Acknowledgments

The research conducted by Jakub M. Tomczak was funded by the European Commission within the Marie Skłodowska-Curie Individual Fellowship (Grant No. 702666, ”Deep learning and Bayesian inference for medical imaging”).