Variational inference over cartesian product of large sets of observations

Questions
5

I do not think I want them in sync. I want it possible for any item in a to pair with any item in b with equal probability. If they are in sync then I think most pairings can never happen.

Or am I thinking about this wrong?