CENS Technical Seminar Series

Learning Low Dimensional Representations of High Dimensional Data

Invited Speaker: Dr. Fei Sha
Date: November 6, 2009
Time: 1:00 PM - 2:00 PM
Venue: BH 4760

Abstract

Statistical modeling of high-dimensional and complex data is a challenging task in machine learning. To tackle this problem, a very powerful strategy is to identify and exploit low-dimensional structures intrinsic to the data. For example, text and image data can often be represented as suppositions of meaningful and interpretable structures such as ``object parts'' and ``topics''. These structures are composed of visually salient image patches as well as groups of semantically related words. Other approaches include manifold learning, where high dimensional data are assumed lying on a low dimensional manifold. These methods for deriving compact representations of high dimensional data have found many applications. They are also relevant and potentially useful for information fusion for sensor networks when we would like to identify meaningful structures from large amount of data collected by sensors.

In this talk, I will give a short survey of some of these methods, focusing on a few research projects I and my collaborators have carried out. I will describe a few interesting applications, one being localizing sensors using pairwise distances.

Biography

Prof. Fei Sha is an assistant professor of the Computer Science Department at the University of Southern California (USC). Prior to that, he earned a Ph.D in Computer and Information Sciences from the University of Pennsylvania. His main research interests and work have been focusing on the theory and application of statistical machine learning. He had earned ``Outstanding Student Paper Award'' at NIPS 2006 and ICML 2004, as well as a finalist for the ``Best Student Paper Award'' at the international conference on acoustics, speech and signal processing (ICASSP, 2007).