Identifier

etd-0327103-141013

Degree

Master of Science in Electrical Engineering (MSEE)

Department

Electrical and Computer Engineering

Document Type

Thesis

Abstract

Neural network architectures that can handle complex inputs, such as backpropagation networks, perceptrons or generalized Hopfield networks, require a large amount of time and resources for the training process. This thesis adapts the time-efficient corner classification approach to train feedforward neural networks to handle complex inputs using prescriptive learning, where the network weights are assigned simply upon examining the inputs. At first a straightforward generalization of the CC4 corner classification algorithm is presented to highlight issues in training complex neural networks. This algorithm performs poorly in a pattern classification experiment and for it to perform well some inputs have to be restricted. This leads to the development of the 3C algorithm, which is the main contribution of the thesis. This algorithm is tested using the pattern classification experiment and the results are found to be quite good. The performance of the two algorithms in time series prediction is illustrated using the Mackey-Glass time series. Quaternary input encoding is used for the pattern classification and the time series prediction experiments since it reduces the network size significantly by cutting down on the number of neurons required at the input layer.

Date

2003

Document Availability at the Time of Submission

Release the entire work immediately for access worldwide.

Committee Chair

Subhash C. Kak

DOI

10.31390/gradschool_theses.3618

Share

COinS