senya-ashukha / semi-supervised-NFs

Code for the paper Semi-Conditional Normalizing Flows for Semi-Supervised Learning

Semi-Supervised Flows PyTorch

Authors: Andrei Atanov, Alexandra Volokhova, Arsenii Ashukha, Ivan Sosnovik, Dmitry Vetrov

This repo contains code for our INNF workshop paper Semi-Conditional Normalizing Flows for Semi-Supervised Learning

Abstract: This paper proposes a semi-conditional normalizing flow model for semi-supervised learning. The model uses both labelled and unlabeled data to learn an explicit model of joint distribution over objects and labels. Semi-conditional architecture of the model allows us to efficiently compute a value and gradients of the marginal likelihood for unlabeled objects. The conditional part of the model is based on a proposed conditional coupling layer. We demonstrate performance of the model for semi-supervised classification problem on different datasets. The model outperforms the baseline approach based on variational auto-encoders on MNIST dataset.

Poster

Semi-Supervised MNIST classification

Train a Semi-Conditional Normalizing Flows on MNIST with 100 labeled examples:

python train-flow-ssl.py --config config.yaml

You can then find logs at <where-script-launched>/logs/exman-train-flow-ssl.py/runs/<id-date>

For the convenience we also provide pretrained weights pretrained/model.torch, use --pretrained flag for loading.

Credits

• Credits to https://github.com/ferrine/exman for the exman parser.

Citation

If you found this code useful please cite our paper

@article{atanov2019semi,
  title={Semi-conditional normalizing flows for semi-supervised learning},
  author={Atanov, Andrei and Volokhova, Alexandra and Ashukha, Arsenii and Sosnovik, Ivan and Vetrov, Dmitry},
  journal={arXiv preprint arXiv:1905.00505},
  year={2019}
}