Physical Society Colloquium

Harnessing the Dynamics of Neural Networks

Larry Abbott

Department of Physiology and Cellular Biophysics
Columbia University

A lot is know about the general dynamic properties of networks of model neurons, and we are beginning to use this knowledge to construct networks that perform tasks and resemble experimental data. Our knowledge of network dynamics is based on mean-field analyses and computations of the spectra of random matrices. I will discuss these methods and the results they provide, and then move on to showing how learning procedures can be used to construct networks with specified dynamic functions.