CAREER: Flexible Statistical Learning for Complex Data

University Of North Carolina At Chapel Hill, Chapel Hill NC

Investigators

Abstract

Statistical learning is widely recognized as a very active area of interdisciplinary research, which lives between statistics, computer science, and optimization. This research offers a host of new statistical learning techniques for solving complicated learning problems, especially for high dimensional and noisy data. In particular, the investigator develops (1). several novel large-margin classifiers which are expected to yield highly competitive classification accuracy, class probability estimation, as well as variable selection; (2). a new regularization approach to estimate the covariance matrix for a class of nonstationary spatial autoregressive model; (3). a novel technique to assess statistical significance of clustering for high dimensional data. With the rapid advance of technology, massive and complex data are being generated across many different scientific fields. Analyzing such data becomes more and more challenging. A major goal of this research is to provide a set of flexible statistical learning techniques. These tools should have beneficial impact on cancer research, medical imaging, microarray data analysis and spatial-temporal modeling. The investigator collaborates with a number of scientists in various fields outside of statistics such as biology, computer science, pharmacy, and genetics. The new developments allow scientists to analyze complex data with high prediction accuracy and increased interpretability. Efficient algorithms and software are developed for public use. The integration of the research goals with educational activities aims to help students at graduate, undergraduate, and high school levels and researchers from various disciplines to acquire state-of-the-art statistical learning methods and tools.