Bayesian spatial modeling with Template Model Builder by Joaquín Cavieres

We are happy to welcome Dr. Joaquín Cavieres (Postdoctoral Fellow @ Bayreuth University) for a presentation on Bayesian spatial modeling using Template Model Builder (TMB). Thanks to Spiced Academy for hosting this meetup!

Speakers:

Dr. Joaquín Cavieres, Postdoctoral Fellow @ Bayreuth University (https://www.joaquincavieres.com/)

Timetable:

19:15 – doors open / socializing
19:30 – welcome
19:45 – talk
20:45 – Q&A
21:00 – socializing
21:30 – end

Title:
Flexible and efficient Bayesian spatial modelling using Template Model Builder (TMB)

Abstract:
Fitting a spatial model using the classic MCMC methods generally is computationally expensive. However, the Hamiltonian Monte Carlo algorithm of Stan offers greater computational efficiency for complex models, such as models with spatial dependence. An alternative to fit spatial models is to use the integrated nested Laplace approximation (INLA) and the stochastic partial differential equation (SPDE) approaches which are implemented in the R-INLA package, which works for a special class of models called “latent Gaussian models”. A part of its computational efficiency is because with the SPDE method we can approximate a Gaussian random field by a Gaussian Markov random field using a discretization of the region of the study. Unfortunately, R-INLA is not a flexible framework to propose a particular likelihood for the response variable or another spatial model different to the approximated GRF.

In this talk, we will see a flexible framework of spatial statistical modelling called Template Model Builder (TMB) and its link with Stan. Furthermore, we will see its computational efficiency under Bayesian inference using the SPDE method and a study case for a Thin Plate Spline spatial model.

Bio:
Joaquín Cavieres is a Postdoctoral Fellow working in the Geoinformatics group at Bayreuth University (Germany). He earned his PhD in Statistics from Universidad de Valparaíso (Chile) and his research is related to spatial modeling, Bayesian inference and computational methods.
For more information visit https://www.joaquincavieres.com/

Note:
Children and (well-behaved) pets are also welcome!