Fast inference for Bayesian ​​nonparametrics

Abstract: Using the symmetry inherent to exchangeable sequences driven by certain classes of Bayesian nonparametric priors, a system of partial difference equations characterising moments of random means is derived and solved. At the outset, the results can be used to compute such quantities in a fast manner, for instance to elucidate Bayesian non-parametric models for a given data set or to attain posterior inferences in a fast manner. However, they can also be used for other inference problems. Some properties and practical examples will be explored.