Date of Award
Electrical and Computer Engineering
Engineering, Electronics and Electrical.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Pattern classification is one of the most successful applications of neural networks. Most of the previous research focused around training the multi-layer perceptrons (MLP's) with the back propagation algorithm. Although the MLP's are capable of resolving pattern classes separated by non-linear class boundaries, they have some drawbacks that limit their acceptance for classifying problems of the real-world. Namely they require very lengthy training times, they cannot learn incrementally, and their convergence is not guaranteed. In this thesis, we develop two prototype based classifiers that require training times that are orders of magnitude less than the MLP's, and can be trained in increments. One of the classifiers uses prototypes of fixed thresholds to represent all classes. The other classifier forms prototypes of different firing thresholds, which it chooses and adjusts according to an algorithm. The classifiers were tested on standard and real world problems like hand-written numeral recognition, in which they demonstrated superior performance in terms of speed of learning and memory efficiency.Dept. of Electrical and Computer Engineering. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis1994 .A28. Source: Dissertation Abstracts International, Volume: 56-01, Section: B, page: 0406. Adviser: Maher Sid-Ahmed. Thesis (Ph.D.)--University of Windsor (Canada), 1994.
Abu-Nasr, Mahmoud A., "On-line fast learning with variable thresholds prototype neural classifier." (1994). Electronic Theses and Dissertations. 2929.