Skip to content

rsaeta/sparse-hyper

 
 

Repository files navigation

Sparse, adaptive hyperlayers

This is the codebase that accompanies the paper Learning sparse transformations through brackpropagation. Follow the link for the paper and an annotated slidedeck.

Disclaimer

We are still cleaning up the code, but it should now be relatively readable. Make sure you have PyTorch 1.0 installed and start by running experiments/identity.py, which runs the identity experiment:

 python experiments/identity.py -F

The -F flag sets all values of the matrix to 1, which makes learning a little easier.

Feel free to ask me for help by making an issue, or sending an email.

The archive branch contains a snapshot of the code at the time the preprint went up.

Dependencies (probably incomplete)

  • Numpy
  • Matplotlib
  • Pytorch 0.4
  • torchvision
  • tensorboardX

About

Code for the paper "Learning sparse transformations through backpropagation"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%