Probabilistic Models with Deep Neural Networks

Recent advances in statistical inference have significantly expanded the toolbox of probabilistic modeling. Historically, modeling has been constrained to very restricted model classes, where exact or approximate is feasible. However, developments variational inference, a general form that originated physics, enabled overcome these limitations: (i) Approximate now possible over broad class models containing large number parameters, and (ii) scalable methods based on stochastic gradient descent distributed computing engines allow be applied massive data sets. One important practical consequence possibility include deep neural networks within models, thereby capturing complex non-linear relationships between random variables. These advances, conjunction with release novel toolboxes, greatly scope applications allowed take advantage recent strides made by learning community. In this paper, we provide an overview main concepts, methods, tools needed use framework.