2 papers across 2 sessions
We develop a highly scalable reparametrization invariant variational training scheme for deep learning.