Multivariate Bayesian Last Layer for Regression: Uncertainty Quantification and Disentanglement - DES Saclay Access content directly
Preprints, Working Papers, ... Year : 2024

Multivariate Bayesian Last Layer for Regression: Uncertainty Quantification and Disentanglement

Abstract

We present new Bayesian Last Layer models in the setting of multivariate regression under heteroscedastic noise, and propose an optimization algorithm for parameter learning. Bayesian Last Layer combines Bayesian modelling of the predictive distribution with neural networks for parameterization of the prior, and has the attractive property of uncertainty quantification with a single forward pass. The proposed framework is capable of disentangling the aleatoric and epistemic uncertainty, and can be used to transfer a canonically trained deep neural network to new data domains with uncertainty-aware capability.
Fichier principal
Vignette du fichier
uai2024_main.pdf (2.31 Mo) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

cea-04567052 , version 1 (02-05-2024)

Licence

Identifiers

  • HAL Id : cea-04567052 , version 1

Cite

Han Wang, Eiji Kawasaki, Guillaume Damblin, Geoffrey Daniel. Multivariate Bayesian Last Layer for Regression: Uncertainty Quantification and Disentanglement. 2024. ⟨cea-04567052⟩
0 View
1 Download

Share

Gmail Mastodon Facebook X LinkedIn More