DeepUncertainty

Documentation for DeepUncertainty.

DeepUncertainty implements techniques generally used to qualtify uncertainty in neural networks. It implements a variety of methods such Monte Carlo Dropout, BatchEnsemble, Bayesian BatchEnsemble, Bayesian Neural Networks. The goal is to have drop-in replacements to convert deterministic layers to Bayesian networks, and also provide examples to help convert custom layers to theor Bayesian counter-parts.

This package is in development as part of the ACED project, funded by ARPA-E DIFFERENTIATE and coordinated by Carnegie Mellon University, in collaboration with Julia Computing, Citrine Informatics, and MIT.