Bayesian Computational Statistics & Modeling
Welcome to Professor Haavard Rue's research group

Page 09/27/2017 17:34:08

Visit and seminar by Prof. Virgilio Gomez-Rubio

9/27/2017


Prof. V. Gomez-Rubio will give a talk "Extending the Integrated Nested Laplace Approximation with
Markov Chain Monte Carlo",  Wednesday, 4 Oct. 2017, 4PM-5PM, B1, L4, seaside near Prof. Rue's office.


Abstract:
The Integrated Nested Laplace Approximation (Rue et al., 2009)
provides a fast and convenient way of obtaining approximations to the
posterior marginals of the model parameters for a large family of models.  In
practice, model fitting is contrained to the latent classes
implemented in the R-INLA package. In this talk, I will describe a new approach
that combines MCMC and INLA to extend the number of models that can
be fitted with R-INLA (Gómez-Rubio and Rue, 2017). Our approach extends
previous work (Bivand et al., 2014) by replacing numerical
approximations on a grid by a sampling method based on the
Metropolis-Hastings algorithm.
During my presentation, I will cover three examples. In the first
one, I will describe how to use priors (univarite or multivariate) not
implemented in R-INLA. In particular, I will describe how to use a
Laplace prior to implement a Bayesian lasso with R-INLA. In the
second example, I will show how to fit models with missing values in the
covariates. Finally, in the third example, I will revisit the models
described in Bivand et al. (2014) to talk about how these and other
spatial models can be fitted with our new method.