Modeling approaches in computational neuroscience generally fall into one of two categories. 1) Mathematical models or 2) Artificial Neural Networks (ANN). Mathematical models use differential equations to describe the state of a neuron, but tend to be computationally expensive and difficult to scale up to large networks of neurons. ANNs model assemblies of elements, but sacrifice features of real brain networks, such as, synaptic transmission, temporal properties and architecture for the ability to build large assemblies. There is a need for a class of models that retain the essential features of single neurons but allow the creation of large networks. We introduce an approach, the Real Time-Qualitative Reasoning Neuron (RT-QRN), to meet this criteria. QR began as an AI technology to describe physical phenomena. The technique is extremely efficient because no precise quantitative values need to be calculated. The precision that mathematical models provide can be unnecessary and burdensome on processing time. Qualitative reasoning focuses on detecting and reporting critical changes and trends in the system under simulation.

The Real Time Qualitative Reasoning Neuron (RT-QRN) is capable of qualitatively reproducing single neuron behavior, but is computationally simple enough to use in large scale neural networks without loss of critical details. Recent QRN simulations of a single Purkinje cell (~1600 compartments) show significant speedup over a recent GENESIS model. A moderately-sized model of the hippocampus (32 neurons, ~700 compartments) is used in an associative learning task and a large scale model of the cerebellar cortex (256 neurons, ~300,000 compartments) is used to simulate a motor control task. 

The QRN Group at the Krasnow Institute for Advanced Studies includes (alphabetical order):  Giorgio Ascoli , Joel Davis, Larry Hunter,  Jeff Krichmar,  Jim Olds and Steve Senft.

Four of the QRN group at a lab meeting.
 
 
 
 

Bibliography:

The Qualitative Reasoning Neuron: A New Approach To Modeling In Computational Neuroscience (Presented at the 6th Annual Computational Neuroscience Meeting, Big Sky Montana)

A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD (Presented at the 27th Annual Meeting of the Society for Neuroscience, New Orleans, LA)

QRN Technical Report 1-1997: A novel and Efficient Approach for Modeling Neural Behavior I: Single Purkinje Cell Model (zipped postscript version)
 

For (p)reprint requests: mailto:jkrichma@osf1.gmu.edu
 

Go to Main Krasnow Institute Home Page

Return to the Computational Neuroanatomy Group