FlowGMM

This repository contains the Julia implementation of the Semi-Supervised Learning with Normalizing Flows paper.

(PyTorch implementation is here)

Downloading the Datasets

Downlaod the files in the following Google Drive links:

Toy Datasets

UCI Datasets (Original UCI datasets: MiniBooNE and HEPMASS. Also the preprocessing from Masked Autoregressive Flow for Density Estimation has been used where sensible )

NLP Datasets (This data has been obtained using this script from the original PyTorch repository. BERT Embeddings of the data has been computed afterwards.)

After downloading the datasets, make three directories named toy_datasets, uci_datasets,and nlp_datasets and move the downloaded datasets to the appropriate folders without any subdirectories.

Results

The experiments are implemented in this notebook.

References

@article{izmailov2019semi,
  title={Semi-Supervised Learning with Normalizing Flows},
  author={Izmailov, Pavel and Kirichenko, Polina and Finzi, Marc and Wilson, Andrew Gordon},
  journal={arXiv preprint arXiv:1912.13025},
  year={2019}
}

@article{dinh2016density,
  title={Density estimation using real nvp},
  author={Dinh, Laurent and Sohl-Dickstein, Jascha and Bengio, Samy},
  journal={arXiv preprint arXiv:1605.08803},
  year={2016}
}