Department Seminar Alexander Stringer Link to join seminar: Hosted on Webex. |
Bayesian Inference for Extended Latent Gaussian Models
I introduce a fast and scalable approximation methodology for posterior distributions in a novel class of Extended Latent Gaussian models. I apply the method in several difficult examples in epidemiology, astrophysics, and other fields, demonstrating the analysis of more complex models and larger data than existing methods. In addition I prove that the approximations yield credible sets with asymptotically correct coverage probabilities. The method improves upon the popular INLA methodology for these reasons. I finish by describing ongoing work including the implementation of the method in open source software.