What is VeBNN?¤

Summary¤

VeBNN provides the implementation for the paper Cooperative variance estimation and Bayesian neural networks disentangle aleatoric and epistemic uncertainties.

Statement of need¤

Real-world data contains aleatoric uncertainty — irreducible noise caused by imperfect measurements or incomplete knowledge of the data-generating process. Mean variance estimation (MVE) networks can learn this type of uncertainty but require ad-hoc regularization strategies to avoid overfitting and are unable to predict epistemic uncertainty (model uncertainty). Conversely, Bayesian neural networks predict epistemic uncertainty but are notoriously difficult to train due to the approximate nature of Bayesian inference. VeBNN introduces a cooperative training strategy between:

a Variance estimation network (aleatoric uncertainty)
a Bayesian neural network (epistemic uncertainty)

They iteratively refine each other, resulting in:

disentangled aleatoric & epistemic uncertainty
improved predictive accuracy
stable training without ad-hoc tricks

A visualization of the VeBNN training procedure is given as follows:

Authorship: - This repo is developed Jiaxiang Yi, a PhD candidate of Delft University of Technology, based on his research context.

Illustrative example

Community Support¤

If you find any issues, bugs or problems with this package, please use the GitHub issue tracker to report them.

License¤

This project is licensed under the BSD 3-Clause License. See LICENSE for the full license text.