Variance Networks

The code for our ICLR 2019 paper on Variance Networks: When Expectation Does Not Meet Your Expectations.

Talk video

Code

We actually have two version of the code:

• TensorFlow implementation is done with python 2.7, and will help to reproduce CIFAR results i.e. training variance layers via variational dropout.
• PyTorch implementation is a way more accurate and reproduces results on MNIST and the toy problem. It requires python 3.6 and pytorch 0.3.

Citation

If you found this code useful please cite our paper

@article{neklyudov2018variance,
  title={Variance Networks: When Expectation Does Not Meet Your Expectations},
  author={Neklyudov, Kirill and Molchanov, Dmitry and Ashukha, Arsenii and Vetrov, Dmitry},
  journal={7th International Conference on Learning Representations},
  year={2019}
}