We develop a highly scalable reparametrization invariant variational training scheme for deep learning.

This paper introduces torch-uncertainty, a unified PyTorch-based framework that benchmarks state-of-the-art uncertainty quantification methods across multiple deep learning tasks and modalities.

Variational Learning Finds Flatter Solutions at the Edge of Stability

A Training-Free Bayesianization approach is proposed for LLM adapters that achieves better uncertainty estimation.

Scalable, post-hoc, Bayesian uncertainty quantification method using an ensemble of linearised networks.