Spring Seminar Series - April 21, 2005
University of Minnesota
School of Statistics
College of Liberal Arts

Techniques for Application-Driven Dimension Reduction of Large Datasets

Anuj Srivastava
Department of Statistics
Florida State University

Thursday, April 21, 2005
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

In many applications dealing with statistical analysis of real data, we are confronted with large data sizes. One often needs to project the data down to smaller dimensions before any statistical modeling or testing can become feasible. This situation arises for example in image analysis, meteorology, and oceanography. Simplicity and efficiency of linear transformations make them a popular tool for reducing dimension before or during statistical analysis. Commonly used techniques are principal component analysis, Fisher's discriminant analysis, independent component analysis, etc. Linear transformations with natural orthogonality constraints can be represented as elements of Stiefel and Grassmann manifolds. We advocate that the choice of a transformation for dimension reduction is not standard; it is dictated by the application and the data set, and can be formulated as an optimization problem on these above-mentioned manifolds. We demonstrate this idea by deriving dimension-reducing transformations in several applications, including image-based recognition of objects and content-based retrieval of images.