1 paper across 1 session
We develop a highly scalable reparametrization invariant variational training scheme for deep learning.