Improving the INLA Approach for Approximate Bayesian Inference for Latent Gaussian Models

• Activity Number: 133
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Egil Ferkingstad* and Håvard Rue
• Companies: Norwegian University of Science and Technology/University of Iceland and Norwegian University of Science and Technology
• Keywords: Bayesian Computation ; INLA ; Approximations ; Copulas

Abstract

Latent Gaussian models is an important and huge class of models, which covers a large part of the statistical models used today.

Integrated Nested Laplace Approximations (INLA) was introduced in 2009 as a tool to do approximate Bayesian inference in these models. The INLA approach has shown both to be very accurate in practice and extremely fast due to the Markov properties of the Gaussian fields used in all the "Laplace" approximations; see www.r-inla.org for software.

In a few cases, there is a little error in these approximations, and in this talk I will report on our experience with a class of improved approximations based on Gaussian copulas, which often do not add any computational costs.