Generalization in multi-layer feed forward nueral networks employing the back propagation algorithm/

Abstract

Neural networks have been around for years, but only recently has there been great interest in them. Neural network models are composed of many computational elements operating in parallel. These elements are called nodes and are nonlinear, typically analog. A type of network which has been used successfully as a classifier is the multi-layer perceptron. Multi-layer perceptrons are feed forward networks with one or more layers of nodes between the input and output nodes. Each node passes the result through a nonlinearity (activation function) to the next layer. This paper uses a neural network as described above to study the capability of the network to generalize. Using English characters as inputs, the network is analyzed for its ability to identify characters that have translation offsets and characters that have been rotated. Also studied is the networks immunity to three cases of noise. In the first case the network is trained with pure characters and noisy characters; then the network's response to characters with an intermediate amount of noise is studied. In the second case the network is trained to pure characters; then the network's response to increasing of noise amounts is observed. Finally an effort is made to train the network with noisy characters, then study the response of the network to noisy characters.

Notes

This item is only available in print in the UCF Libraries. If this is your thesis or dissertation, you can help us make it available online for use by researchers around the world by downloading and filling out the Internet Distribution Consent Agreement. You may also contact the project coordinator Kerri Bottorff for more information.

Graduation Date

1989

Semester

Summer

Advisor

Georgiopoulos, Michael

Degree

Master of Science (M.S.)

College

College of Engineering

Department

Electrical Engineering and Communication Sciences

Format

PDF

Pages

28 p.

Language

English

Length of Campus-only Access

None

Access Status

Masters Thesis (Open Access)

Location

Orlando (Main) Campus

Identifier

DP0026688

Subjects

Dissertations, Academic -- Engineering; Engineering -- Dissertations, Academic

Accessibility Status

Searchable text

This document is currently not available here.

Share

COinS