**10/12/08 - 14:00h** Ciclo de Palestras Pós Graduação em Estatística UFRJ

["Alexandra M. Schmidt" <alex@im.ufrj.br>]

Caros, 

Encerrando as atividades  do Ciclo de Palestras do Programa de Pós-Graduação 
em Estatística do IM-UFRJ em 2008, teremos, 

nesta **4a feira, 10/12/08**, excepcionalmente as **14:00h** as seguintes 
palestras 

Palestrante: Leonardo S. Bastos (University of Sheffield, UK) 
Título: Validating Gaussian process emulators

Palestrante: Thais C. O. Fonseca (University of Warwick, UK) 
Título: Nongaussian spatiotemporal modeling  

Local: Sala C-116, Centro de Tecnologia, Cidade Universitária.

Os resumos seguem abaixo.

Aproveito a oportunidade para desejar aos membros da lista boas festas e um 
ano novo de muita saude e paz!


Abracos
Alexandra

Ps.: Desculpem-me pela eventual duplicacao desta mensagem.



Resumos:

Título: Validating Gaussian process emulators

In this talk, I am going to give a brief introduction of complex Computer 
models, also known as simulators, that are widely used in all areas of 
science and technology to represent complex real-world phenomena. Simulators 
are often sufficiently complex that they take appreciable amounts of 
computer time or other resources to run. In this context, a methodology has 
been developed based on building a statistical representation of the 
simulator, known as an emulator. The principal approach to building 
emulators uses Gaussian processes. I am going to present a set of 
diagnostics to validate and assess the adequacy of a Gaussian process 
emulator as surrogate for the simulator. These diagnostics are based on 
comparisons between simulator outputs and Gaussian process emulator outputs 
for some test data, known as validation data, defined by a sample of 
simulator runs not used to build the emulator. Our diagnostics take care to 
account for correlation between the validation data.   
 

Título: Nongaussian spatiotemporal modeling

In this work, we develop and study nongaussian models for processes that 
vary continuously in space and time. The main goal is to consider heavy 
tailed processes that can accommodate both aberrant observations and 
clustered regions with larger observational variability. These situations 
are quite common in meteorological applications where outliers are 
associated with severe weather events such as tornados and hurricanes. In 
this context, the idea of scale mixing a gaussian process as proposed in 
Palacios and Steel (JASA, 2006) is extended and the properties of the 
resulting process are discussed. The model is very flexible and it is able 
to capture variability across time that differs according to spatial 
locations and variability across space that differs in time. This is 
illustrated by an application to maximum temperature data in the Spanish 
Basque Country. The model allows for prediction in space-time since we can 
easily predict the mixing process and conditional on the latter the finite 
dimensional distributions are gaussian. The predictive ability is measured 
through proper scoring rules such as log predictive scores and interval 
scores. In addition, we explore the performance of the proposed model under 
departures from gaussianity in a simulated study where data sets were 
contaminated by outliers in several ways; overall, the nongaussian models 
recover the covariance structure well whereas the covariance structure 
estimated by the gaussian model is very influenced by the contamination.   
 

-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Alexandra Mello Schmidt, PhD
Professora Adjunta
Instituto de Matemática - UFRJ
Departamento de Métodos Estatísticos
Caixa Postal 68530 Rio de Janeiro - RJ 
CEP:21.945-970 Brasil
Tel: 0055 21 2562 7505 Ramal (Extension) 204
Fax: 0055 21 2562 7374

http://www.dme.ufrj.br/~alex
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
There is no time to lose. Cash your dreams before they slip away. 
Lose your dreams and you lose your mind.
Gentileza gera gentileza (Profeta Gentileza).