Feature Extraction vs. Flexible Modeling: Automated Modeling of Complex, Irregular Object Data

• Activity Number: 372
• Type: Invited
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Jeffrey S Morris*+
• Companies: MD Anderson Cancer Center
• Keywords: Functional data analysis ; Image analysis ; Object data ; Nonparametric models ; Bayesian modeling ; High dimensional data

Abstract

Our modern society has developed numerous new measurement instruments producing an explosion of data, ever-growing in their quantity and complexity. One emerging type is object data, which have inherent structure that, when taken into account, can lead to improved inference. They pose the challenge of how to efficiently and reliably extract the valuable scientific information they contain while managing the practical challenges raised by their size and subtleties. The absence of sufficiently flexible models can force researchers to first reduce their data using simple summaries to eliminate some of their vexing complexities. This can work well if the summaries retain all relevant information in the data, but many times that is not the case. One primary objective of modern statistics is to develop efficient, flexible methods that can model the complex data as they are, while avoiding oversimplifications, and thereby potentially uncovering more of the treasure trove of information they contain. I illustrate these principles through several real-world applications, and summarize some work I have done developing flexible methods and general frameworks for analyzing object data.