April 22-24, 2014 - room G5-112 (each day 13.00-16.00)
Finn Lindgren, University of Bath and Daniel Simpson, Norwegian University of Science and Technology, Trondheim
will give lectures on SPDE/GMRF models and associated computational methods.
Combining computational methods for Gaussian Markov random fields with continuous spatial domain models (Lindgren et al., 2011, JRSSB) has generated a lot of interest in applied statistics for spatial mapping in epidemiology and ecology, as well as in general environmetrics and geography. In these six lectures we will discuss the theory and practice for spatial modelling and inference using stochastic partial differential equations and finite basis representations.
Basic lecture plan:
Gaussian Markov random fields (GMRF); theory and spatial statistics
Stochastic partial differential equations (SPDE); continuous theory and finite approximations
Integrated Nested Laplace Approximation (INLA); fast, direct, Bayesian inference
Point processes; log-Gaussian Cox-processes based on SPDE/GMRF models
Non-stationarity, space-time, and joint excursions; interpretation, parameterisation, and computation
Big computations and numerical methods; computing large log-determinants, and other problems for big models
Host: Jesper Møller