Uncertainty quantification

A novel method for Bayesian inference in deep ensembles using Stein Variational Newton updates, leveraging scalable Hessian approximations to improve uncertainty quantification, training efficiency, and robustness.