Masters Theses

Date of Award

12-1990

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Bruce J. MacLennan

Abstract

The purpose of this paper is to examine the issues of network representation and methods of referring to elements of artificial neural networks as they are implemented in three publicly available simulators for such networks. It will be shown that these issues are central to the performance of these simulators, and have other implications as well. A critical comparison of other features of these simulators will also be offered. The three simulators used in the study are; the Rochester Connectionist Simulator (RCS) from the University of Rochester, the California Institute of Technology's GENESIS simulator, and the SFINX simulator from the University of California at Los Angeles. The question of network representation arises when a neural network is prepared for simulation. It involves precisely how the network should be described for thew simulator, and what data structures will be used by the simulator to carry out the simulation. A related question involves how the user will refer to the elements in the network in controlling the simulation. The approach taken will be to provide some background material about each of the simulators under discussion, their history and development, methods of network representation and reference, and their basic resource requirements. Then a test simulation designed to establish the simulator's response to networks of increasing size will be described and its implementation on the simulators will be discussed. Finally, the results of the tests will be examined. It is hoped that this study may be of interest those who must choose a simulator (either from the public or commercial market places) for use in an educational/pedagogic or research setting.

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

Share

COinS