Uncertainty-based OOD detection

Exploring the link between uncertainty estimates obtained by "exact" Bayesian inference and out-of-distribution (OOD) detection.

Prerequisits

The code in this repository requires the installation of the hypnettorch package.

Experiments and example usage

The subfolder notebooks jupyter notebook to reproduce experiments from our papers, but they also show how to use the code in this repo. Further usage examples can be found in the subfolder tutorials.

Neural network Gaussian process

The folder nngp contains utilities to work with NNGP kernels.

Documentation

Documentation can be found in folder docs. Using sphinx, the documentation can be compiled within this folder by executing make html. The compiled documentation can be opened via the file index.html.

Citation

When using this package in your research project, please consider citing one of our papers for which this package has been developed.

@inproceedings{henning:dangelo:2021:bayesian:ood,
title={Are Bayesian neural networks intrinsically good at out-of-distribution detection?},
author={Christian Henning and Francesco D'Angelo and Benjamin F. Grewe},
booktitle={ICML Workshop on Uncertainty and Robustness in Deep Learning},
year={2021},
url={https://arxiv.org/abs/2107.12248}
}