Doctoral Dissertations

Tzusheng Pei

Date of Award

12-1999

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Computer Science

Major Professor

Bruce MacLennan

Committee Members

Jim Drake, Jim Hall, Michael Vose

Abstract

The goal of this research is to investigate to what degree randona artificial dendritic nets can differentiate between temporal patterns after modifying the synaptic weights of certain synapses according to a learning algorithm based on the Fourier transform.

A dendritic net is organized into subnets, which provide impulse responses to a function as a basis for Fourier decomposition of the input pattern. Each subnet is randomly generated. According to the simulations, randomly generated subnets with appropriate parameters are good enough to provide the impulse responses for the Fourier decomposition.

The electrical potential pattern across the membrane of the dendrites follows the cable equation. The simulations use a linear synapse model, which is an approximation to biologically realistic synapses. Both excitatory and inhibitory synapses are present in a dendritic net.

The simulations show that random dendritic nets with a small number of subnets can be modified to differentiate between electrical current patterns to a high degree when the membrane conductance of the dendrites is high, and they also show that the random structures are highly fault-tolerant. The performance of a random dendritic net does not change much after adding or deleting subnets.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS