All You Need is a Good Functional Prior for Bayesian Deep Learning

Code for the paper "All You Need is a Good Functional Prior for Bayesian Deep Learning".

The code is under refactoring, feel free to contact me via email (ba-hien.tran@eurecom.fr) if you have any issues or questions.

Setup

We assume python3.6 or python3.7 since the package was developed with those versions. To install the package with pip3 please clone this repository and then run the following

pip3 install .

Example Usage

The subfolder notebooks contains jupyter notebooks to run experiments with regression and classification tasks. They also show how to use this package. Here, we included some demos as follows

• 1D_regression_Gaussian_prior.ipynb: Comparison between FG and GPiG priors on a 1D regression data.
• 1D_regression_hierarchical_prior.ipynb: Comparison between FH and GPiH priors on a 1D regression data set.
• 1D_regression_norm_flow_prior.ipynb: Comparison between Fixed NF and GPiNF priors on 1D regression data.
• 2D_classification.ipynb: The effect of using different configurations of the target GP prior to the predictive posterior on a 2D classification task.
• 2D_classification_hierarchical_gp_prior.ipynb: The effect of using a target hierarchical-GP prior to the predictive posterior on a 2D classification task.
• uci_regression.ipynb: Comparison between FG and GPiG priors on a UCI data set.

Citation

When using this package in your work, please consider citing our paper

@article{Tran2022,
  author    = {Tran, Ba-Hien and Rossi, Simone and Milios, Dimitrios and Filippone, Maurizio},
  title   = {{All You Need is a Good Functional Prior for Bayesian Deep Learning}},
  journal = {Journal of Machine Learning Research},
  volume  = {23},
  pages   = {1--56},
  year    = {2022}
}