We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Department of Mathematical Sciences

Seminar Archives

On this page you can find information about seminars in this and previous academic years, where available on the database.


Presented by Janine Illian , University of St Andrews

20 October 2014 02:05 in CM221

Integrated nested Laplace approximation (INLA) may be used to fit a large class of (complex) statistical
models. While MCMC methods use stochastic simulations for estimation, integrated nested Laplace ap-
proximation (INLA) is based on deterministic approximations where there are no convergence issues. INLA
is a very accurate and computationally superior alternative to MCMC and may be used to fit a large class of
models, latent Gaussian models.
Since INLA is fast, complex modelling has become greatly facilitated and has also become more accessible
to non-specialists. In addition, due to the fact that the fitting approach is embedded in a large and general
class of statistical models, very general types of models may be considered. This allows us a lot more
flexibility in the choice of model than previously – and hence the models to capture interesting aspects of
the data and consequently the system they are relevant for. In the context of spatial statistics, for example, we
can now fit models to spatial point patterns of high dimensionality, replicated point patterns, hierarchically
marked point patterns etc. In many cases, analysing these data sets with MCMC approaches would be very
cumbersome and computationally prohibitive.
The INLA-methodology has been implemented in C, and the associated numerical calculations and algo-
rithms rely on an efficient implementation of numerical procedures for Gaussian Markov random fields
(GMRF), in particular the algorithms in the C-library GMRFLib. However, most users do not need to worry
about this, as the INLA-methodology has been made accessible through a user-friendly R-library, R-INLA,
described and available for download at Specifying and fitting models using R-INLA
is just as easy as applying standard routines in R, for example fitting generalised linear models, and it also
provides great flexibility with regard to the models that may be fitted.
In order to illustrate INLA's versatility I will discuss a range of examples and present a number of recent
developments. This concerns generalisations of the methodology, functionality within the R-INLA library
and discussions on prior choice.

Contact for more information