Wide Bayesian neural networks have a simple weight posterior: theory and accelerated sampling.

Jiri Hron, Roman Novak, Jeffrey Pennington, Jascha Sohl-Dickstein

Browse the full ICML paper archive.