Digital implementation of back propagation neural networks

Abstract

A neural network is a powerful tool that is able to capture and represent complex input-to-output relationships. The motivation to develop such technologies originated from the desire to develop an artificial system that could perform "intelligent tasks" similar to those performed by the human brain. An Artificial neural network (ANN) is basically an interconnected group of artificial neurons that uses a mathematical or computational model in order to process information. In most cases, ANN is an adaptive system that changes its structure based on the internal or external network information. In 1969, Minsky and Papert showed that there are many simple problems such as the exclusive-or problem which linear neural networks cannot solve or learn. If such simple problems cannot be solved, how could they solve other complex problems such as vision, language, and motor control. One of the solutions to this problem is the Back Propagation Learning Algorithm. The back propagation algorithm was developed by Paul Werbos in 1974 and was rediscovered independently by Rumelhart and Parker. The back propagation algorithm is the learning algorithm popularly used in feed forward multilayer neural networks. The main objective of this dissertation was to implement the various components of the Back Propagation Neural Network into many individual modules, using VHSIC hardware description language (VHDL). Each module was tested individually and various simulation results were achieved. Afterwards, each module was connected in a top level architecture in order to form part of a structural neural network model.