Signal and Information Processing in Analysis and Modeling of Complex Structures

点击翻译按钮获取中文摘要

Computational technology advances in storing and processing large amounts of complex-structured data have inevitably created a demand for fast and reliable transformation of data to accurate information about its underlying structure. The structures of interest range from man-made ones such as modern energy infrastructures or hyperspectral images, to living cells, organs of the human body or speech. These structures are modeled mainly for the purpose of analysis, prediction, and/or control of their behavior. All of these purposes heavily rely on the accuracy of the modeling process, which itself depends on our prior knowledge of the system and the data available from it. Proposing practical methods for data-driven modeling as well as efficient schemes for evaluating the models are critical challenges. These issues become vital especially in modeling health-related systems and structures where use of an unreliable model that can not be validated may cause irreversible damages.The intended outcome of this research is to provide knowledge, theory, and algorithms that can be applied to optimize performance and cost efficiency of data-driven structural modeling. The focus of this proposal is developments in two specific areas: (1) system structure data-driven modeling, parametric complexity, order selection, and model evaluation for linear and nonlinear systems; and (2) learning from data with a concentration on data denoising, classification, and compression, with a focus on Hyperspectral Imaging (HSI). This research program will benefit industries involved with data analysis and structural modeling. Immediate impact will be seen in the area of HSI by providing reliable and fast analysis, especially for the new application areas of medical HSI.