Incorporating Unlabelled Data into Bayesian Neural Networks

Mrinank Sharma, Tom Rainforth, Yee Whye Teh, Vincent Fortuin

Published 2023-04-04Version 1

We develop a contrastive framework for learning better prior distributions for Bayesian Neural Networks (BNNs) using unlabelled data. With this framework, we propose a practical BNN algorithm that offers the label-efficiency of self-supervised learning and the principled uncertainty estimates of Bayesian methods. Finally, we demonstrate the advantages of our approach for data-efficient learning in semi-supervised and low-budget active learning problems.