Skip to content

Low-variance, efficient and unbiased gradient estimation for optimizing models with binary latent variables. (ICLR 2019)

License

Notifications You must be signed in to change notification settings

mingzhang-yin/ARM-gradient

Repository files navigation

ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks

Code to show the simulation results in ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks

Data sets

The MNIST data is self-contained and the Omniglot data is in the repository.

Citations

Below are the paper to cite if you find the algorithms in this repository useful in your own research:

@inproceedings{yin2018arm,
title={{ARM}: Augment-{REINFORCE}-Merge Gradient for Stochastic Binary Networks},
author={Mingzhang Yin and Mingyuan Zhou},
booktitle={International Conference on Learning Representations},
year={2019}
}

License Info

This code is offered under the MIT License.

About

Low-variance, efficient and unbiased gradient estimation for optimizing models with binary latent variables. (ICLR 2019)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages