Design for novel enhanced weightless neural network and multi-classifier.

Weightless neural systems have often struggles in terms of speed, performances, and memory issues. There is also lack of sufficient interfacing of weightless neural systems to others systems. Addressing these issues motivates and forms the aims and objectives of this thesis. In addressing these issues, algorithms are formulated, classifiers, and multi-classifiers are designed, and hardware design of classifier are also reported. Specifically, the purpose of this thesis is to report on the algorithms and designs of weightless neural systems.

A background material for the research is a weightless neural network known as Probabilistic Convergent Network (PCN). By introducing two new and different interfacing method, the word "Enhanced" is added to PCN thereby giving it the name Enhanced Probabilistic Convergent Network (EPCN). To solve the problem of speed and performances when large-class databases are employed in data analysis, multi-classifiers are designed whose composition vary depending on problem complexity. It also leads to the introduction of a novel gating function with application of EPCN as an intelligent combiner. For databases which are not very large, single classifiers suffices. Speed and ease of application in adverse condition were considered as improvement which has led to the design of EPCN in hardware. A novel hashing function is implemented and tested on hardware-based EPCN.

Results obtained have indicated the utility of employing weightless neural systems. The results obtained also indicate significant new possible areas of application of weightless neural systems.