Critical Branching Neural Computation

by Chris Kello

University of California, Merced

Date

Friday, September 18, 2009

Time

3:00 p.m. — 4:00 p.m.

Place

Lindley Hall, Rm. 102

Abstract: Brain networks transmit and process information via action potentials (i.e. spikes) generated by neurons and transmitted via synapses. Spiking activity, as quantified by sums of spikes over neurons, cannot overly expand or contract over time. Stability is achieved in the balance, and can be expressed in terms of critical branching. I present a spiking neural network model with a local tuning algorithm that converges spiking dynamics to their critical branching point. The model is used as a liquid state machine to examine its computational capacity as defined by the effects of perturbations (e.g. environmental inputs) to spiking dynamics. The model exhibits maximal capacity near the critical branching point, as well as power law distributions and fluctuations in spiking activity. These power laws are observed in various measures of neural and behavioral activities, suggesting that critical branching may indicate and illustrate a general principle of cognition.

Biography: Christopher Kello is an Associate Professor of Cognitive Science at the University of California, Merced. His career in cognitive science began with a B.A. at University of Rochester, Ph.D. at UC Santa Cruz, Postdoc at Carnegie Mellon, Faculty at George Mason University, and Program Director at the National Science Foundation. His work began with neural network models of reading and speech, coupled with experimental work in word reading. While this work continues, Dr. Kello recently began studying scaling laws in languages and human behavior as evidence for a relation between criticality and cognitive processing.

Colloquium Provided By:

the School of Informatics