Abstract
Probabilistic deep learning is deep learning that accounts for uncertainty,both model uncertainty and data uncertainty. It is based on the use ofprobabilistic models and deep neural networks. We distinguish two approaches toprobabilistic deep learning: probabilistic neural networks and deepprobabilistic models. The former employs deep neural networks that utilizeprobabilistic layers which can represent and process uncertainty; the latteruses probabilistic models that incorporate deep neural network components whichcapture complex non-linear stochastic relationships between the randomvariables. We discuss some major examples of each approach including Bayesianneural networks and mixture density networks (for probabilistic neuralnetworks), and variational autoencoders, deep Gaussian processes and deep mixedeffects models (for deep probabilistic models). TensorFlow Probability is alibrary for probabilistic modeling and inference which can be used for bothapproaches of probabilistic deep learning. We include its code examples forillustration.