https://github.com/mirceamironenco/BayesianRecurrentNN

Implementation of Bayesian Recurrent Neural Networks by Fortunato et. al
https://github.com/mirceamironenco/BayesianRecurrentNN

bayesian-inference deep-learning machine-learning

Last synced: 11 months ago
JSON representation

Implementation of Bayesian Recurrent Neural Networks by Fortunato et. al

• Host: GitHub
• URL: https://github.com/mirceamironenco/BayesianRecurrentNN
• Owner: mirceamironenco
• License: mit
• Created: 2017-10-20T12:13:01.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2018-12-17T07:08:39.000Z (over 7 years ago)
• Last Synced: 2024-11-27T04:31:00.094Z (over 1 year ago)
• Topics: bayesian-inference, deep-learning, machine-learning
• Language: Python
• Homepage:
• Size: 11.7 KB
• Stars: 217
• Watchers: 8
• Forks: 40
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Bayesian Recurrent Neural Networks

This is a replication of the paper 'Bayesian Recurrent Neural Networks' by Meire Fortunato, Charles Blundell, Oriol Vinyals.

Paper: https://arxiv.org/abs/1704.02798

Status: Basic model replicated.

# Requirements

- Python 3.5

- [TensorFlow 1.3.0](https://www.tensorflow.org/)

# Usage

    $ sh download_ptb.sh

    $ python bayesian_rnn.py -model medium -log_sigma1 -1.0 -log_sigma2 -7.0 -prior_pi 0.25

### To-do:

- Implement posterior sharpening.

- Implement image captioning experiment.

### Acknowledgements

Thanks to Meire Fortunato for providing the Bayes by Backprop/cell code and @alexkrk for an initial implementation.