Functional Variational Bayesian Neural Networks

This code is jointly contributed by Shengyang Sun, Guodong Zhang and Jiaxin Shi.

Introduction

Code for "Functional variational Bayesian neural networks" (https://arxiv.org/abs/1903.05779)

Dependencies

This project runs with Python 3.6. Before running the code, you have to install

• Tensorflow
• GPflow-Slim

Experiments

Periodic Prior | RBF Prior :-------------------------:|:-------------------------: |

Below we shows some examples to run the experiments.

x3 regression

python exp/toy.py -d x3 -in 0.01

sinusoidal extrapolation

python exp/toy.py -d sin -na 40 -nh 5 -nu 500 -e 50000 -il -2

Inference on Implicit Piecewise Priors

python exp/piecewise.py -d p_const

Regression

python exp/regression.py -d yacht

Contextual Bandits

python exp/bandits.py --data_type statlog

Citation

To cite this work, please use

@article{sun2019functional,
  title={Functional Variational Bayesian Neural Networks},
  author={Sun, Shengyang and Zhang, Guodong and Shi, Jiaxin and Grosse, Roger},
  journal={arXiv preprint arXiv:1903.05779},
  year={2019}
}