1) Bayesian deep learning combines deep learning and Bayesian modeling to address some limitations of each approach. It allows for principled uncertainty quantification in predictions and can model non-stationarity. 2) Deep learning performs well but only provides point estimates without uncertainty. Bayesian modeling provides uncertainty in predictions but has seen little application to machine learning. 3) Bayesian deep learning uses probabilistic programming to specify models with priors and perform inference to obtain posterior distributions over weights, enabling uncertainty estimates in deep learning.