Nested Semi-Definite Cone Analysis with Application to Diffusion Tensor Image Data

• Activity Number: 225
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Lingsong Zhang*+ and Sungkyu Jung
• Companies: Purdue University and University of Pittsburgh
• Keywords: Nested Cone Analysis ; Principal Component Analysis ; Diffusion Tensor Image ; Object-oriented data analysis ; Subspace learning ; Nonnegative matrix factorization

Abstract

Motivated by Diffusion Tensor Imaging, we propose a nested semi-definite cone analysis, which provides a series of approximations of different ranks to the original data. At each rank k, all of the approximations lie in a dimension k subspace, and also are semi-definite, which leads to better interpretation, compared to other existing methods. Extensive simulations will be used to compare the connections and differences between our method and existing methods. The merit of our method will be shown in the application of this method to a Diffusion Tensor Image data set.