• Fall 2009
• Spring 2010
• Fall 2010
• Spring 2011
• Fall 2011

Seminar Series: Spring 2012
The seminar series will usually take place on Fridays in a Business Building room, but the exact time and location could be different due to a variety of factors including room availability in the Business Building. For actual direction to the Business Building, please see campus map. For additional information contact Dr. Kefeng Xu, (210) 458-5388.

Friday, Jan. 20, 2012, 10:30-11:50am, Business Building 4.02.10 (Executive Conference Room), joint seminar with ASA-San Antonio Chapter.

• Presenter: Robert McCulloch, Professor of Information, Risk & Operations Management Department, McCombs School of Business, University of Texas at Austin

• Joint Work with Hugh Chipman (Acadia), Ed George (U Pennsylvania),
  James Gattiker, Dave Higdon, Matt Pratola, Los Alamos National Laboratories

• Presentation Title: Bayesian Additive Regression Trees and Parallel Computation

• Abstract: In BART: Bayesian Additive Regression Trees, we developed a general approach for fitting the model y = f(x) + e with minimal assumptions about the structure of f. The function f is modeled as a sum of simple tree models.  This sum is made meaningful by a Bayesian analysis with a simple yet powerful prior specification that provides for both overall regularization of the fit and the limiting of the contribution of each individual tree to the sum.   An MCMC algorithm provides inference. The BART model is shown to have excellent out of sample predictive performance.
  In this talk we review BART and report results on an implementation using MPI (Message Passing Interface) for parallel computation. For problems with large sample sizes,  the algorithm is k times faster when k processors are used.

Friday, Feb. 10, 2012, 12noon-1pm, Business Building 4.02.10 (Executive Conference Room).

• Presenter: Joel Edmund Michalek, PhD Professor and Vice Chairman, Department of Epidemiology & Biostatistics University of Texas Health Science Center at San Antonio

• Co-authors: Ken Ouyang MS, and Chris Louden MS

• Presentation Title: A Fourier and Wavelet Analysis of Serial Glucose Measurements

• Abstract: Normoglycemia maintained by tight glucose control protocols are reported to reduce morbidity and mortality in critically ill surgical patients. Glucose variability has been used to characterize glycemic control and efficacy of tight glucose control protocols; however, the definition and methods for determining glucose variability have not been standardized. This was a single-center retrospective review of adult trauma patients admitted to the Surgical Trauma Intensive Care Unit at a Level I Trauma Center between January 2005 and December 2007. Patients were managed with a tight glucose control protocol with target range 80-110mg/dl. Subjects were identified by trauma registry query and glucose values were obtained through the hospital database. Patients with an intensive care unit length of stay <3 days and those with mild injury severity were excluded. Glucose variability was calculated using glucose measurements made during the first 3 days using two measures of glucose curve complexity, Fourier power and Wavelet energy. The primary endpoint was death between day 3 and day 30. Logistic regression and Cox models were used to assess the significance of associations between complexity and death and between complexity and time to death. Associations found were significant but not predictive.

   

Friday, March 23rd, 2012, 2-3pm, Business Building 4.02.10 (Executive Conference Room).

• Presenter: Long Liu, PhD, Assistant Professor, Department of Economics, University of Texas at San Antonio

• Presentation Title: The Estimation and Testing of a Linear Regression with Near Unit Root in the Spatial Autoregressive Error Term

• Abstract: This paper considers the estimation of a linear regression involving the spatial autoregressive (SAR) error term which is nearly nonstationary. The asymptotics properties of the ordinary least squares (OLS), true generalized least squares (GLS) and feasible generalized least squares (FGLS) estimators as well as the corresponding Wald test statistics are derived. The limiting behavior of spatial regression diagnostics such as Moran I and Lagrange Multiplier (LM) statistics are also analyzed. Monte Carlo results are conducted to study the sampling behavior of the proposed estimators and test statistics.

Monday, April 10, 2012. 10:00-11:30 am, Stat Lab (BB 3.02.16).

• Presenter: Jose A. Bird, Ph.D. Candidate, Ph. D. APPLIED STATISTICS DISSERTATION DEFENSE, Department of Management Science & Statistics, University of Texas at San Antonio

• Presentation Title: COMBINING DIVERSE CLASSIFIERS USING CLASS SPECIFIC PRECISION INDEX FUNCTION

• Abstract: In this work the overall precision index (PIN) (Ko and Windle 2011) is extended to a class-specific precision index function (PIC) and used to combine diverse classifiers. The PIN is a measure of overall local prediction accuracy for a classifier, whereas the PIC is a measure of local class prediction accuracy. The main motivation for extending the precision index function to be class-specific is to generate higher precision prediction sets than can be used to identify potential candidates such as when classifying proteins to sub-cellular location sites. This work compares the performances of PIC and PIN to other combining methods including majority voting, stacked generalization, and cluster-selection for the well-known data sets: 1) vowel recognition data (Hastie, Tibshirani et al. 2009) which is a balanced data set, and 2) yeast protein localization data (Frank 2010) which is an unbalanced data set. When comparing the PIC method to other combining methods for the vowel recognition data, the PIC method was not able to generate high precision prediction sets. Similar results were obtained with an extension of the static cluster-selection method to the class-specific level. Modified PIC curves were generated which used the two classes predicted by a classifier with the highest posterior probabilities for each prediction point. This enhancement increased the overall precision of the PIC method and extended the results to higher precisions for the vowel recognition data. A new weighted precision index from PIC and PIN was also developed which further extended the PIC results to higher precisions. The weighted precision index method outperformed most combining methods and slightly outperformed the PIN method at higher precisions. Even though the PIC method was outperformed by other methods at lower precisions for the yeast localization data, it generated higher recalls in the high precision range. The class-specific cluster selection extension, considered in this work, outperformed all other combining methods for the yeast localization data set demonstrating great potential for this method to leverage class-specific performance. The overall precision results obtained with the yeast protein localization data set for both precision index methods and cluster-selection methods outperformed previous results reported by Chen (Chen 2010) where several classifying methods including: decision trees, neural networks, naives Bayes, and Bayesian model averaging methods were considered.

Friday, April 20, 2012, 10:00-11:00am, Business Building 4.02.10 (Executive Conference Room).

• Presenter: Angel Díaz Matalobos, PhD, Professor of Operations and Director of PhD Program, Instituto de Empresa Business School, María de Molina 12 bajo, Madrid 28006, Spain

• Presentation Title: To Be Announced (related to Supply Chain Management)

• Abstract: T.B.A.

•  

Friday, April 27, 2012, 2-3:15am, Business Building 4.02.10 (Executive Conference Room).

• Presenter: Kail Li, PhD, Former Dean of Management School, Professor of Operations Management, and Director of Institute for Industrial Economics and Management, Northeastern University, China.

• Presentation Title: To Be Announced (related to Supply Chain Management)

• Abstract: T.B.A.