Proposal: DMS 95-04414 PI(s): Jianqing Fan - 1989, Stephen Marron - 1982 Institution: U of NC Chapel Hill Title: Processing Massive Noisy Data with Hidden Structure Abstract: This research involves a few inter-related statistical problems ranging from processing images and finding informative structure from massive data, to analyzing survival data, evaluating the success of a business advertisement and understanding the home range of animals. In contrast with traditional approaches, the methods do not assume any restrictive form, and are flexible enough to detect fine structures. This project covers a wide array of statistics methodological developments and foundational research. The new developments enhance the availability of statistical tools, and can lead to significant improvements on existing methods for extracting information from noisy data. Specific objectives are develop new statistical methodologies and investigate their foundational properties for flexible statistical modeling in processing high-dimensional data, nonparametric confidence intervals, mode detection and signal processing. Many scientific disciplines depend in some way on extracting structural information from noisy data. Fields ranging from processing noisy images and evaluating business marketing, to analyzing survival data and forecasting economic climates, which are universes apart in their backgrounds, nevertheless have the common problem of drawing conclusions via processing noisy signals. Such problems may be abstracted as a statistical function estimation problem and can be analyzed by various techniques in this project. The objective of this research is to develop and evaluate flexible statistical modeling techniques. These techniques can be applied in the Federal Strategic Areas such as monitoring environmental and global changes via processing massive collected data where informative structures can hardly be detected by traditional approache s, and building statistical modeling for economic and business activities in the civil infrastructure. The investigators will take advantage of modern computing facilities, and use statistical knowledge to avoid unnecessary data mining and hence reduce significantly data processing time. Thus, the knowledge gained in this study will also be useful in high performance computing of a federal strategic area.
