Rubén Zamar
Rubén Zamar - University of British Columbia (CANADÁ)
Ruben H. Zamar graduated as a Public Accountant from University of Cordoba, Argentina in 1973. He received his M.Sc. in Statistics from the Inter-American Teaching Center of Statistics, OAS, Chile, in 1977, his M.Sc. in Mathematics from Universidade Federal de Pernambuco, Brazil, in 1981 and his Ph.D. degree in Statistics from University of Washington, U.S.A. in 1985.
Ruben Zamar is Professor of Statistics at the University of British Columbia, Canada since 1986. He has served as Associate Editor for several journals includ-ing the Annals of Statistics, the Journal of the American Statistical Association and Test. He published over 60 articles in statistical journals and conferences.
Ruben Zamar has supervised or co-supervised ten Ph.D students and many M.Sc. students. His research interest includes robustness, computational statistics, data mining, image processing and cluster analysis.
Research stay at UC3M: DEPARTMENT OF STATISTICS
Project: I am looking forward to my future collaboration with professors Peña, Romo, Berrendero and Dr. Viladomat. Peña, Viladomad and Zamar (2012) proposes a novel approach to cluster analysis based on repeated multivariate medians of near-neighboring points. We proved some nice mathematical properties – such as the convergence to attracting points located near modes - for the case of one-dimensional data. We now wish to extend our mathematical results to the multi-dimensional setting. Moreover, we wish to produce a “scalable” version of the algorithm that can be applied to very large datasets. Prof. Peña and I also share a common interest in Bayesian robustness. We share the belief that Bayesian methods should be robust against data contamination as well as changes to the prior distributions. To address this issue Peña et al (2009) develope a robust Bayesian approach to linear regression, which is resistant to outliers and high leverage points. We wish to extend our robust Bayesian approach to other models including multivariate location and scatter, non-linear regression and multivariate time series analysis.
Prof. Romo and I are interested in the problem of sparse clustering of datasets when there is a very large number of variables, possibly exceeding the number of cases. In this setting it is reasonable and convenient to select a relatively small number of variables to conduct the clustering. Following Witten and Tibshirani (2010) we believe that the selection of the clustering variables can be better achieve when performed simultaneously with the clustering procedure itself. Outliers and other types of contamination are often present in large datasets. We wish to develop robust and sparse clustering procedures that can deal with outliers, missing data and scale-up with the number of variables and the number of cases.
Stay Period: OCT 12 - APR 13
Conferences
Lecturer: Profesor Dr. Ruben Zamar
Title: A natural robustification of the ordinary instrumental variables estimator
Date: March 15 at 13:00h
Place: Aula 17.2.75 (Getafe Campus)
