Sampling for Inference in Probabilistic Models with Fast Bayesian Quadrature

The central challenge in probabilistic inference is numerical integration, to average over ensembles of models or unknown (hyper-)parameters (for example to compute the marginal likelihood or a partition function).