Uncertainty estimation under model misspecification in neural network regression

Analysis of the implications of model misspecification for estimating uncertainty when doing regression when neural networks.

Requirements

You need to install the hypnettorch library by doing python3 -m pip install hypnettorch.

Documentation

How to build the documentation is explained in the docs folder.

Running experiments

For running experiments, move to the regression_uncertainty_mm subfolder. Further instructions can be found on the README there.

Citation

When using this package in your research project, please consider citing our workshop paper for which this repository has been developed.

@inproceedings{cervera2021regression,
title={Uncertainty estimation under model misspecification in neural network regression},
author={Maria R. Cervera and Rafael Dätwyler and Francesco D'Angelo and Hamza Keurti and Benjamin F. Grewe and Christian Henning},
booktitle={NeurIPS Workshop on Robustness and misspecification in probabilistic modeling},
year={2021},
}