AI SearchThe back-propagation scheme in data classification : software and empirical guidelines
Juneja, Hursh
The back-propagation scheme in data classification : software and empirical guidelines - Bangkok : Asian Institute of Technology, 1992 - 105 leaves - Thesis ; no. CS-92-10 . - Asian Institute of Technology. Thesis ; no. CS-92-10 .
A thesis submitted in partial fulfillment of the requirement for the degree of Master of Engineering, School of Engineering and Technology
Thesis (M.Eng.) - Asian Institute of Technology, 1992
The present study explores the applicability of Neural Networks to the problem of classification which is of great importance in many areas. For this purpose, a common neural network with back propagation aigorithm was used. First of all, a fully menu-driven and user friendly software package was developed to facilitate the application of this type of networks. It caters to numerical data input and is a fas~ high precision tool for neural network training. From a careful comparative study, it was found that for both normal and non-normal data, neural networks can result in better classification accuracy as comped to classical statistical methods, the latter being based on the assumption of an underlying data distribution. An attempt has also been made to formulate some guieielines to answer the questions regarding selection of a suitable learning rate and changing of thi ~ rate for faster convergence while training without any modifications to the training software. A third guideline has been proposed to decide the extent of training itself. All these can be achieved by careful examination of error and actual output curves of the training.
Neural networks (Computer science)
Information storage and retrieval systems
The back-propagation scheme in data classification : software and empirical guidelines - Bangkok : Asian Institute of Technology, 1992 - 105 leaves - Thesis ; no. CS-92-10 . - Asian Institute of Technology. Thesis ; no. CS-92-10 .
A thesis submitted in partial fulfillment of the requirement for the degree of Master of Engineering, School of Engineering and Technology
Thesis (M.Eng.) - Asian Institute of Technology, 1992
The present study explores the applicability of Neural Networks to the problem of classification which is of great importance in many areas. For this purpose, a common neural network with back propagation aigorithm was used. First of all, a fully menu-driven and user friendly software package was developed to facilitate the application of this type of networks. It caters to numerical data input and is a fas~ high precision tool for neural network training. From a careful comparative study, it was found that for both normal and non-normal data, neural networks can result in better classification accuracy as comped to classical statistical methods, the latter being based on the assumption of an underlying data distribution. An attempt has also been made to formulate some guieielines to answer the questions regarding selection of a suitable learning rate and changing of thi ~ rate for faster convergence while training without any modifications to the training software. A third guideline has been proposed to decide the extent of training itself. All these can be achieved by careful examination of error and actual output curves of the training.
Neural networks (Computer science)
Information storage and retrieval systems