Seminartio/UNICAMP

[Ronaldo Dias <dias@ime.unicamp.br>]

Prezados(as)

   Nesta quinta 13/05 as 14:30hs sala 221, teremos a presenca do Prof. 
Marco Ferreira/UFRJ apresentando o seguinte seminario:

SEMINÁRIO DE ESTATÍSTICA E PROBABILIDADE

Marco A. R. Ferreira
DME/UFRJ

Bayesian Analysis for a Class of Gaussian Markov Random Fields

Resumo: We present an objective Bayesian analysis for a classof proper 
Gaussian Markov random fields (PGMRF). These proper PGMRFs are obtained 
by adding a positive constant called propriety parameter to the diagonal 
of the improper GMRF precision matrix. The propriety of GMRFs is 
important because it is necessary for the construction of more complex 
models such as for example multi-resolution random fields and 
spatio-temporal models.Here, we show that the difficulties reported by 
some authors onthe Bayesian estimation of the propriety parameter result 
from the behavior of the likelihood function when the propriety 
parameter goes to infinity. In particular, this behavior implies that 
the use of improper priors leads to improper posteriors. In order to 
solve this problem, we develop an objective Bayesian analysis for the 
parameters of PGMRFs. We prove that our corresponding reference priors 
lead to proper posteriors. Moreover, we have found through simulation 
studies that our proposed approach has very good frequentist properties. 
Illustrations are given with simulated data.

Data: 13/05/2004 Hora:  14:30hs.    Local:  Sala de Reuniões do  Depto. 
de Estat. 221.


Contamos com sua presenca.


Ronaldo

-- 
Ronaldo Dias
Depto. Estatistica, IMECC - UNICAMP
dias@ime.unicamp.br
www.ime.unicamp.br/~dias