Network Information Theory and Computational Neuroscience

Professor Todd Coleman, Coordinated Science Laboratory, Electrical and Computer Engineering, University of Illinois at Urbana-Champaign
This talk discusses two topics. First, we consider a network information theory problem of communicating correlated sources over broadcast networks in a distributed fashion. We discuss the breakdown of Shannon's celebrated source-channel separation theorem for this class of problems. We next characterize the full capacity region and illustrate that a certain desirable architecture is optimal: source-channel separation at the encoders and joint-source channel decoding. We next discuss the implications and directions for further research.

Secondly, we switch to inference methods for estimating dynamical systems in computational neuroscience. We consider characterizing the learning of an unfamiliar task with simultaneous behavioral and neurophysiological responses. Specifically, we track the cognitive state of a monkey as it learns a location-scene association task by combining the measurements of its correct and incorrect intra-trial responses, its intra-trial reaction times, and its inter-trial neural spiking activity from cortex. We cast the problem in terms of developing practical methodologies for performing state estimation on linear stochastic dynamical systems from simultaneous measurements of different types (continuous, binary, and point process) over different time scales. We develop recursive filtering and smoothing techniques which as special cases reduce to the Kalman filter. We develop an EM algorithm to fit the model with experimental data according to maximum likelihood. With this approach we are able to develop goodness-of-fit tests to asses the fit of the data with the model.

Since July 1, 2006, Todd P. Coleman has been an Assistant Professor at UIUC in the Department of Electrical and Computer Engineering and is also affiliated with the Neuroscience Program. His research interests are in network information theory, wireless communication, operations research, and computational neuroscience. Todd received B.S. degrees in electrical engineering as well as computer engineering from the University of Michigan in 2000. He completed the M.S. degree in 2002 and Ph.D. degree in November 2005, both in electrical engineering from MIT, under the supervision of Prof. Muriel Medard. For the remaining 2005-2006 academic year, Todd was a postdoctoral scholar in the Neuroscience Statistics Research Laboratory at MIT's Department of Brain and Coginitive Sciences and Massachusetts General Hospital, under the supervision of Prof. Emery Brown, M.D., Ph.D.