Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness

Bayesian neural networks and deep ensembles are principled approaches to estimate the predictive uncertainty of a deep learning model.