arXiv 2015

Variational Dropout and the Local Reparameterization Trick

arXiv 2015 JavierAntoran/Bayesian-Neural-Networks

Our method allows inference of more flexibly parameterized posteriors; specifically, we propose variational dropout, a generalization of Gaussian dropout where the dropout rates are learned, often leading to better models.

BAYESIAN INFERENCE