CNS Seminar

I'll talk about a current research focus in my laboratory: Mapping arbitrary nonlinear dynamical systems onto networks of spiking neurons. The Neural Engineering Framework (NEF) is a formal method for accomplishing this mapping that is robust to heterogeneity as well as noise in the neural population—it actually exploits the former to approximate any desired mathematical function. Using NEF, you can, for example, get a spiking-neuron network to perform temporal integration or rotate an n-dimensional vector by a given angle. Why would you use NEF instead of the standard paradigm for training neural networks? The resulting interconnection-weight matrix has low rank and thus can be compressed, reducing storage requirements by over an order of magnitude. And why would you use spiking neurons rather than non-spiking ones (i.e., rate neurons)? Spiking neurons can be implemented energy-efficiently with subthreshold analog circuits and interconnected in a reconfigurable fashion using asynchronous digital circuits.

I'll present a Kalman-filter-based brain-machine interface and a three-degree-of-freedom robot-arm controller mapped onto a mixed analog-digital neuromorphic chip using NEF. These results demonstrate that NEF is robust to heterogeneity introduced by transistor mismatch as well as noise introduced by independent spike arrival times (Poisson stochastic process). Suggesting that, combining analog computation with digital communication in this fashion may effectively combat increased mismatch and noise as transistors scale down to a few nanometers. Intriguingly, the brain adopts precisely such a hybrid approach to combat its nanoscale ion-channels' stochastic expression and operation: Its neurons' dendritic trees use graded potentials to combine thousands of inputs while their axonal arbors use all-or-none spikes to transmit the resulting output to thousands of other neurons.

Bio: Kwabena Boahen received the B.S. and M.S.E. degrees in electrical and computer engineering from the Johns Hopkins University, Baltimore, MD, both in 1989 and the Ph.D. degree in computation and neural systems from the California Institute of Technology, Pasadena, CA, in 1997. He was on the bioengineering faculty of the University of Pennsylvania from 1997 to 2005, where he held the first Skirkanich Term Junior Chair. He is presently a Professor in the Bioengineering Department of Stanford University, with a courtesy appointment in Electrical Engineering. He directs Stanford's Brains in Silicon Laboratory, which develops silicon integrated circuits that emulate the way neurons compute, linking the seemingly disparate fields of electronics and computer science with neurobiology and medicine.

Prof. Boahen's contributions to the field of neuromorphic engineering include a silicon retina that could be used to give the blind sight, a self-organizing chip that emulates the way the developing brain wires itself up, and a specialized hardware platform (Neurogrid) that simulates a million cortical neurons in real-time—rivaling a supercomputer while consuming only a few watts. He has received several distinguished honors, including a Fellowship from the Packard Foundation (1999), a CAREER award from the National Science Foundation (2001), a Young Investigator Award from the Office of Naval Research (2002), a Pioneer Award from the National Institutes of Health (2006), and a Transformative Research Award from the National Institutes of Health (2011). His 2007 TED talk, "A computer that works like the brain", has been viewed half-a-million times.