SUPR
Bayesian uncertainty quantification for neural networks
Dnr:

NAISS 2024/22-1119

Type:

NAISS Small Compute

Principal Investigator:

Hampus Linander

Affiliation:

Göteborgs universitet

Start Date:

2024-08-30

End Date:

2025-09-01

Primary Classification:

10799: Other Natural Sciences not elsewhere specified

Webpage:

Allocation

Abstract

Proper modelling of uncertainty for neural networks is an active field of research with many applications, both from a theoretical and practical perspective. Bayesian modelling provides a theoretically grounded framework to reason about uncertainty in a systematic fashion, and it's specific application to neural networks is referred to as Bayesian Neural Networks. This projects aims to implement new methods to approximate the true Bayesian posterior distribution that the theory provides but often is computationally intractable.