Serra do Curral   

Professor,  Departament  of  Computer  Science

Universidade Federal de Minas Gerais, Belo Horizonte, Brazil

Research Areas :  Spatial Statistics and Machine Learning

I am a Professor in the Department of Computer Science, Universidade Federal de Minas Gerais (UFMG) located in Belo Horizonte, Brazil. I received my Ph.D. in Statistics in 1994 from the University of Washington, Seattle, USA, having Peter Guttorp as advisor. From 1994 to 2011, I worked in the Department of Statistics, UFMG. I moved to my current position in September 2011. I prefer to work in applied problems where the stochastic nature of the problem requires probabilistic and statistical modeling. My main application areas have been epidemiology social issues, actuarial risk, and demographic analysis. Due to my recent move to the computer science department, I have been adding new application areas to my work. This interdisciplinary work in turn shape my methodological research in statistics and machine learning. 

My current research is focused on the development of new algorithms and statistical methods to analyse spatial and space-time data. I am primarily concerned with the spatial analysis of risk appearing in many fields such as epidemiological surveillance, geosensor networks, enviornmental problems, marketing, spatially variable risk spatially variable, among many others. The computer revolution, still going on, have made it possible to manage huge space-time databases. The extraction of interesting patterns from these massive databases create new, challenging, and interesting problems that require creative algorithmic, statistical and probabilistic solutions.

Posted on February 06, 2014 by Renato  

Bayesian spatial models with a mixture neighborhood structure 


Recent work with Erica Rodrigues, a faculty at the Universidade Federal de Ouro Preto, appeared recently in the Journal of Multivariate Analysis. In Bayesian disease mapping, one needs to specify a neighborhood structure to make inference about the underlying geographical relative risks. We propose a model in which the neighborhood structure is part of the parameter space. We retain the Markov property of the typical Bayesian spatial models: given the neighborhood graph, disease rates follow a conditional autoregressive model. However, the neighborhood graph itself is a parameter that also needs to be estimated. We investigate the theoretical properties of our model. In particular, we investigate carefully the prior and posterior covariance matrix induced by this random neighborhood structure, providing interpretation for each element ofthese matrices.

Posted on February 06, 2013 by Renato  

Optimal Generalized Truncated Sequential Monte Carlo Test


This is another paper just accepted (Feb 2013) at the Journal of Multivariate Analysis, a joint work with Ivair Ramos Silva, based on his PhD dissertation at UFMG. The abstract reads as follows:
Conventional Monte Carlo tests require the simulation of $m$ independent
copies from the test statistic U under the null hypothesis $H_0$. The execution
time of these procedures can be substantially reduced by sequentially
monitoring the simulations. The literature has evaluated the properties
of specific sequential Monte Carlo designs to perform hypothesis tests implementations
by using restrictions on the probability distribution of
the p-value. Such restrictions are used to bound both the resampling risk,
the probability that the accept/reject decision is different from the decision
from the exact test, and the expected execution time. The application of
the main proposals in the literature depends on specific algorithms and its
power for finite number of simulations were not explored by its authors.
This paper develops a generalized sequential Monte Carlo test that includes
the main previous proposals and that allows an analytical treatment of the
power and the expected execution time. These results are valid for any test statistic.
We define the sequential risk, the probability that the accept/reject decision is
different from the decision from the conventional Monte Carlo test, and construct
an optimal sequential procedure which minimizes the expected number of simulations
within a large set of designs. We also bound the resampling risk by consider
 a large class of p-value distributions.

The Self-Feeding Process: A Unifying Model for Communication Dynamics in the Web

My second submission to a CS conference.
Accepted at the 23rd International World-Wide Web Conference (WWW 2013).
Pedro Olmo Vaz de Melo, Christos Faloutsos, Renato Assuncao and Antonio Loureiro.