GPytorch with Lightning in the Lightning-UQ-Box

[nilsleh]

We are excited to share the first release of the Lightning-UQ-Box, a PyTorch domain library using Lightning and GPyTorch to implement various different UQ-Methods for Deep Learning models. Our background is Earth Observation data, however, we hope that this can be an open-source effort to bridge communities and make UQ more accessible to practitioners and possibly even enhance new research avenues.

At the moment, we have implemented Deep Kernel Learning (DKL) and Deterministic Uncertainty Estimation (DUE) based on the input we received in #1759 , however, there of course exist numerous ways to improve and extent the combination of GPyTorch and Lightning. We are looking for feedback, suggestions, and of course possible contributions, so if you have any questions, feel free to reach out :)

Repo: https://github.com/lightning-uq-box/lightning-uq-box
Documentation: https://lightning-uq-box.readthedocs.io/en/latest/
DKL Tutorial: https://lightning-uq-box.readthedocs.io/en/latest/tutorials/regression/deep_kernel_learning.html