arXiv:2405.01761v1 Announce Sort: new
Summary: We current new Bayesian Final Layer fashions within the setting of multivariate regression underneath heteroscedastic noise, and suggest an optimization algorithm for parameter studying. Bayesian Final Layer combines Bayesian modelling of the predictive distribution with neural networks for parameterization of the prior, and has the enticing property of uncertainty quantification with a single ahead cross. The proposed framework is able to disentangling the aleatoric and epistemic uncertainty, and can be utilized to switch a canonically skilled deep neural community to new information domains with uncertainty-aware functionality.
Supply hyperlink