Research Abstracts Online
January 2010 - March 2011
University of Minnesota Twin Cities
College of Liberal Arts
PI: Yuhong Yang
Error Rate Estimation and Multi-Armed Bandit Problems
With the availability of high-dimensional data in medical research, business and finance, engineering, and more, statistical tools have been quickly developed to handle large datasets in terms of the number of predictors and/or the number of observations. High-dimensional classification is an important statistical problem. This project explores new methods to estimate error rate in classification and compare their performances with bootstrap and other proposals in the literature in terms of interval estimation. Another direction is multi-armed bandit problems with covariates, which have applications in sequential clinical trials, project scheduling, resource allocation and experimentation, and others. The researchers are working on methods to effectively and efficiently estimate the mean reward functions of the different arms and ways to select exploration probability. Different numerical methods will be compared to search for the best choice of arms at the different covariate values.
Gang Cheng, Graduate Student