A pytorch implementation of the Bayesian Neural Network, posterior inference is based on Bayes-by-backprop method.
I found some pre-existing implementation alas with bugs and ambiguity. I tried to include explaination on the places I feel confused about first and figured out later. The notebook is tested bug-free from my jupyter notebook.
Many thanks to Javier Antoran from whose code I learned a lot.
@fangyuan-ksgk