deep gaussian processes (DGPs) are hierarchical generalizations of Gaussian Pro- cesses that combine well calibrated uncertainty estimates with the high flexibility of multilayer models. One of the biggest challenges with these models is that exact →