Dr. Alyson Fletcher
Monday, May 21, 2012, 10:00 AM to 11:30 AM
Location: Engineering 2, Room 599
Hosted By Ken Pedrotti

Abstract

The staggering size and complexity of problems from biological systems to communication networks coupled with increases in computing power ushered in a new era of interest and impact in large-scale data analysis and dimension reduction.  This rapidly developing area lies in the intersection of mathematics, machine learning, and signal processing. Sparsity-based methods and graphical models play a growing role.

Answering the questions of how low-dimensional structures can be tractably inferred from data, how current methods perform, and how to improve algorithmic design is important as we move forward.

First, the recovery of a sparse signal from noisy random linear measurements is considered.  Via tools from statistical physics and information theory, sharp limits for reliable recovery are derived for a number of methods including Lasso and matching pursuit.  The role of key data parameters are identified and utilized in improved algorithm selection and design.  New message passing methods incorporating even richer model classes are presented, and their performance is characterized. These underlying mathematical and algorithmic structures have strikingly wide applicability. Connections to problems in wireless random access and estimation in neuroscience are shown.

Biography

Allie Fletcher received the M.S. in mathematics and the M.S. and Ph.D. in electrical engineering from the University of California, Berkeley. She is a recipient of a UC President's Postdoctoral Fellowship, the UC Berkeley EECS Lawler Award, a Luce Foundation fellowship, and an NSF Graduate Fellowship. Her research interests include information theory, statistical inference, optimization, computational neuroscience, biological imaging, and epilepsy seizure detection.