Utility of hypernetworks for optimal transport

[nathanieljevans]

Given two distributions, P and Q, the goal of optimal transport is to learn a transport function that transforms P to Q, typically with some cost constraint.

Given our stochastic prediction, I think we should be able to do this conveniently with simple objectives. To be seen.