back Back to all Seminars

Statistics Seminar: Approximate posterior inference for latent Gaussian models with a multivariate link function

Start Date: February 26, 2019
End Date: February 26, 2019

By Prof. Birgir Hrafnkelsson, University of Iceland
Latent Gaussian models (LGMs) form a frequently used class within Bayesian hierarchical models. This class is such that the density of the observed data conditional on the latent parameters can be any parametric density, and the prior density of the latent parameters is Gaussian. Typically, the link function is univariate, i.e., it is only a function of the location parameter. Here the focus is on LGMs with a multivariate link function, e.g., LGMs structured such that the K parameters in the data density of each observation are transformed to K latent parameters. These K latent parameters are modeled with a linear model at the latent level. The parameters within the linear model are also defined as latent parameters and thus assigned a Gaussian prior density. To facilitate fast posterior computation, a Gaussian approximation is proposed for the likelihood function of the parameters. This approximation, along with a priori assumption of Gaussian latent parameters, allows for straightforward sampling from the posterior density. The computational approach is demonstrated on; (i) data for fitting regression lines on multiple grid points; (ii) annual maximum peak flow series from UK.
Biography: Birgir Hrafnkelsson is a Professor of Statistics in Department of Mathematics at University of Iceland. Department of Mathematics is within the Faculty of Physical Sciences, which is under the School of Engineering and Natural Sciences at UoI. He finished a Ph.D. degree in Statistics in 1999 from Texas A&M University. His Ph.D. thesis was on multiple time series and his advisor was Prof. H. Joseph Newton. Prior to that he completed a four year long C.S. degree and a M.Sc. degree in Mechanical Engineering from the University of Iceland in 1993 and 1995. After finishing his PhD program, he was a Postdoctoral Researcher in the Department of Statistics at The Ohio State University where he worked under the direction of Prof. Noel Cressie. During this time Birgir was introduced to spatial statistics, Markov random fields and Bayesian hierarchical models among other things. Birgir moved to Iceland in 2000. He was a Postdoctoral Researcher in the Modelling Department at the Marine Research Institute in Iceland from 2000 to 2001, where he worked under the direction of Prof. Gunnar Stefánsson. From 2001 to 2003 he was a Researcher at deCODE genetics. Birgir has been at University of Iceland since 2003. He became a Professor of Statistics at University of Iceland in 2016. His research interests include Bayesian hierarchical modeling, spatial statistics, spatio-temporal modeling, extreme value analysis and statistical modeling of environmental data. His current research projects involve computation for latent Gaussian models, spatio-temporal modeling of glaciology data with inclusion of PDEs, statistical modeling of earthquake engineering data and spatial modeling of flood data. Birgir is an Honorary Visiting Professor within College of Engineering, Mathematics and Physical Sciences at University of Exeter from 2017.

More Information:

For more info contact: Prof. Raphael Huser : email:
Date: Tuesday 26th Feb 2019
Time:04:00 PM - 05:00 PM
Location: Building 1, Level 4, room# 4102
 Light refreshments will be served around 3:45 PM