Friday, October 19, 2012 - 11:45am to 1:00pm
Christopher J. Rozell, Ph.D., Assistant Professor, Electrical and Computer Engineering, Georgia Institute of Technology
Christopher John Rozell, Ph.D., Assistant Professor, Electrical and Computer Engineering, Georgia Institute of Technology. Title: Dynamical systems for modeling, measurement and inference with sparse signals. In this talk I will give an overview of some recent results where modeling, measurement and inference for sparse signals intersects with the field of dynamical systems. The first area of focus in the talk will be on dynamic signal models, where temporal regularity can be used to improve estimates of sparse coefficients that change with time. I will introduce an algorithm based on re-weighted L1 minimization that extends the notion of Kalman filtering to a robust and efficient causal estimator for dynamic sparse signals. The second area of focus in the talk will be on dynamical systems for performing inference in sparse signal models. I will describe our current work studying and implementing a continuous time dynamical system that performs inference much faster and with lower power than algorithms running on a digital computer. Time permitting, I will also briefly highlight our work analyzing performance when the measurement process itself is a dynamical system, leading to a network model of sequence memory.