Also nobody fits neural networks and use variation inference using any priors that aren’t some standard form that makes algorithm easy