Author

Franz

Date of Award

12-1998

Degree Name

Master of Science

Department

Computer Science

First Advisor

Dr. Robert Trenary

Second Advisor

Dr. Ajay Gupta

Third Advisor

Dr. John Kapenga

Access Setting

Masters Thesis-Open Access

Abstract

A portable, object-oriented library for simulation of general Multi-layer Feedforward Neural Networks (MLFNs) is described. Unlike all-encompassing neural network simulation environments, the library was designed to allow convenient use in existing programs and in applications where training and testing data are generated using separate, often complex simulations.

The library' s design goals include modularity, portability, efficiency, correctness, compactness, and type-safety. To demonstrate how these objectives are met, competing architectural choices are presented, along with the criteria used for determining the strategy actually implemented. Sample applications using the library are presented, showing how the library' s class files are used in neural network simulations. Finally, the performance of the library is evaluated to demonstrate that the neural network algorithms chosen exhibit modest run-time and storage costs.

Share

COinS