O Departamento de Estatística - UFMG está organizando um ciclo de seminários
nesta sexta-feira, dia 07 de março, de 14:00 às 17:00.
Teremos a participação do Prof. Peter Mueller, do Anderson Cancer Center, do Prof. Luigi Ippoliti, da Universitá "G d'Annunzio" e do Prof. Marcelo Azevedo Costa, da UFMG.
Seguem abaixo os horários das palestras, que irão ocorrer na sala de seminários (2076) do ICEx-UFMG:
14:00 - Peter Mueller
14:50 - Luigi Ippoliti
15:40 - café
16:00 - Marcelo Azevedo
O resumo dos seminários são apresentados abaixo:
1) Bayesian Clustering with Regression
Peter Mueller, M.D. Anderson Cancer Center
We propose a model for covariate-dependent clustering, i.e., we
develop a probability model for random partitions that is indexed by
covariates. The motivating application is inference for a clinical
trial. As part of the desired inference we wish to define clusters
of patients. Defining a prior probability model for cluster memberships
should include a regression on patient baseline covariates. We build
on product partition models (PPM). We define an extension of the PPM
to include the desired regression. This is achieved by including in
the cohesion function a new factor that increases the probability of
experimental units with similar covariates to be included in the same
cluster.
We discuss implementations suitable for continuous, categorical,
count and ordinal covariates.
2) Interpolating spatial gaussian random fields by gaussian
Markov random fields
Luigi Ippoliti, Dipartamento di Metodi Quantitativi e Teoria
Economica - Universitá "G d'Annunzio"
The talk considers the role played by Gaussian Markov Random Fields – GMRFs –
(conditional autoregression models) in interpolating covariance-based
Gaussian Random Field models (Geostatistics models). Both parametric and non-
parametric methods for estimating the GMRF parameters are considered. A
measure of linear interpolability of the process is also constructed by
considering an index of linear determinism, AF , which gives a measure of
the lack of explanation in the linear predictor. An application of this
index for model specification and diagnostic testing of a Gaussian Markov
Random Field is also investigated
3) Optimizing Monte Carlo simulations in Scan Statistics
Marcelo Azevedo Costa - Departamento de Estatística - UFMG
Suppose we observe a number of points located within a rectangular geographical or spatial area. These points may for example reflect the locations of trees, ant nests, diseased individuals or post offices. The general aim of the spatial scan statistic is to detect and evaluate the statistical significance of a spatial or space-time cluster events that cannot be explained by an underlying probability model defining the null hypothesis. The complexity of this problem lies in the multiple testing inherent in the many window locations and the overlapping nature of those windows, resulting in the maximum being taken from a set of highly dependent observations. The challenge lies when the null distribution of the scan statistic is not known or hard to obtain. In this situation it might be possible to generate samples from it. This study aims at optimizing the number of samples since large samples increase computational cost and provide p-values with better resolution. However, techniques as sequential Monte Carlo, extreme distributions and static Monte Carlo provide reliable inference results using just a fraction of conventional compute time.
Atenciosamente,
Glaura Franco