Radial Bayesian Neural Networks

Description

This repository contains the code for the paper Radial Bayesian Neural Networks: Beyond Discrete Support in Large-Scale Bayesian Deep Learning.

We only run experiments on the MNIST dataset.

How to run

First, install dependencies

# clone src   
git clone https://github.com/RomanShen/radial-bnn.git

# install dependencies 
cd radial-bnn
pip install -r requirements.txt

Next, run either convolutional or radial version MNIST experiments.

# convolutional version
python run_conv.py    

For multiple runs with different seeds, go to WandB Sweeps for help.

Basically, run following commands for convolutional version.

wandb sweep sweep_conv.yaml
wandb agent your-sweep-id

Results

All experimental results are available online here.

Citation

@InProceedings{pmlr-v108-farquhar20a, 
title = {Radial Bayesian Neural Networks: Beyond Discrete Support In Large-Scale Bayesian Deep Learning}, 
author = {Farquhar, Sebastian and Osborne, Michael A. and Gal, Yarin}, 
booktitle = {Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics}, 
pages = {1352--1362}, 
year = {2020}, 
editor = {Silvia Chiappa and Roberto Calandra}, 
volume = {108}, 
series = {Proceedings of Machine Learning Research}, 
month = {26--28 Aug}, 
publisher = {PMLR}, 
pdf = {http://proceedings.mlr.press/v108/farquhar20a/farquhar20a.pdf}, 
url = { http://proceedings.mlr.press/v108/farquhar20a.html },