Lab Home | Phone | Search
Center for Nonlinear Studies  Center for Nonlinear Studies
 Colloquia Archive 
 Postdoc Seminars Archive 
 Quantum Lunch 
 Quantum Lunch Archive 
 CMS Colloquia 
 Q-Mat Seminars 
 Q-Mat Seminars Archive 
 P/T Colloquia 
 Kac Lectures 
 Kac Fellows 
 Dist. Quant. Lecture 
 Ulam Scholar 
 CNLS Fellowship Application 
 Student Program 
 Past Visitors 
 History of CNLS 
 Maps, Directions 
 CNLS Office 
Wednesday, January 21, 2009
3:00 PM - 4:00 PM
CNLS Conference Room (TA-3, Bldg 1690)


Stochastic Learning in Brains and Machines

Todd Leen
Oregon Health & Sciences University

Over the last decade neurophysiologists have found that learning-related synaptic changes in several systems are sensitive to the relative timing between pre- and post- synaptic spike events. The temporal resolution involved is on a scale of a few tens of milleseconds. This spike-timing-dependent plasticity (STDP) stands in contrast to earlier experimental results that lacked this time resolution, and hence described synaptic plasticity in terms of correlations between pre- and post-synaptic events estimated over many spikes.

Rate-based learning models, involving statistical averages, are conveniently described by deterministic differential or difference equations. STDP learning models, on the other hand, have intrinsic random components, and their dynamics are properly described in terms of stochastic dynamical systems. Synaptic strengths evolving under STDP and the iterative statistical estimators in stochastic approximation algorithms from machine learning share a common dynamical description in terms of a, usually intractable, master equation and it approximations.

I will discuss viable and non-viable approximate solution techniques for the ensemble dynamics, and an exact solution for a particular STDP learning rule observed in rat hippocampal neurons.

Host: Andy Fraser, Space & Remote Sensing, ISR-2