Friday, March 2, 2012 at 3:00 PM
8 St. Mary’s Street, Room 203
Refreshments served at 2:45.
Efficiently Learning Sparse Representations
Solving a lasso problem is a popular approach to finding a sparse representation of a new item of data with respect to a dictionary of atoms or codewords. Such representations lie at the heart of many recent machine learning and statistics applications. Given a large set of training data one can go one step further and learn a dictionary that gives sparse representations of the training data. However, learning a large scale dictionary from high dimensional training data can quickly become computationally onerous. Two approaches appear promising for making this task more efficient: building a dictionary from small components that can be easily learnt, for example in a tree structured fashion, and reducing large sparse representation problems to smaller ones. In the latter case, one uses the test data to quickly ascertain those codewords in the dictionary that with some level of confidence are not important for test data under study. We discuss some recent advances on these problems.
Peter Ramadge received the B.Sc., B.E. and the M.E. degree from the University of Newcastle, Australia, and the Ph.D. degree from the Department of Electrical Engineering at the University of Toronto, Canada. He joined the faculty of Princeton University in September 1984, where he is currently Gordon Y.S. Wu Professor of Engineering, Professor of Electrical Engineering and Chair of the Department of Electrical Engineering.
He is a Fellow of the IEEE and a member of SIAM. He has received several honors and awards including: a paper selected for inclusion in IEEE book “Control Theory: Twenty Five Seminal Papers (1932-1981)”; an Outstanding Paper Award from the Control Systems Society of the IEEE; the Convocation Medal for Professional Excellence from the University of Newcastle, Australia; an Engineering Council Teaching Award from the School of Engineering and Applied Science, Princeton University; an IBM Faculty Development Award; and the University Medal from Newcastle University, Australia.
His current research interests are in high dimensional signal processing, fMRI imaging, medical imaging, and video/image processing. This includes projects concerning multi-subject functional registration, classification from fMRI datasets, and machine learning.
For the past six years he has been engaged in active collaborations with several neuroscientists at Princeton and elsewhere on applying ideas from signal processing and machine learning to problems in fMRI data analysis.
Hosting Professor: Venkatesh Saligrama
Student Host: Yanfeng Geng