Stein's Method in Computational Statistics

Dr Chris Oates, University of Newcastle. Part of the Statistics seminar series.

This talk will overview an exciting topic in computation for Bayesian statistics, that offers an alternative to the Monte Carlo methods which are widely used. It will be shown how the judicious use of Stein's method enables both optimisation-based alternatives to sampling and super-efficient approximation of posterior integrals defined on a Riemannian manifold.