Tilbake til søkeresultatene

BEMATA-Beregningsorientert matematikk i anv.

Evaluation of Bayesian hierarchical models

Tildelt: kr 3,8 mill.

A trend in statistical analysis has been to consider more complex models. Such highly structured stochastic systems require numerical algorithms for making inference, which is a field of extensive research. Model evaluation in this context has however not been seriously addressed. One reason is the formidable computational challenges involved. This project aims to develop tools for model evaluation in complex models, with special emphasis on Bayesian hierarchical models. Evaluation of the appropriateness of a model must involve what the aims of the constructed models are. In some substance sciences, the models themselves are the main interest, and are selected and validated to obtain improved scientific understanding. In other situations the models are primarily tools for estimating quantities of interest, and it is the model's ability to f.ex. predict that has to be evaluated. In this project we shall be concerned with both global and focussed measures, and although our goal is to invent general tools, we will concentrate on some selected applications in telemobile communications and fisheries research. In the comparison of models, computational efficiency is crucial. Development of faster Markov chain Monte Carlo (MCMC) algorithms will be addressed with special emphasis on adaptive MCMC.

Budsjettformål:

BEMATA-Beregningsorientert matematikk i anv.

Temaer og emner

Ingen temaer knyttet til prosjektet