Real valued classification using complex neural networks

• Main Author: Pushkar Shukla.
• Other Authors: School of Computer Engineering
• Format: Final Year Project
• Language: English
• Published: 2012
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: http://hdl.handle.net/10356/48587
• Institution: Nanyang Technological University

This report details the conception, design , implementation and analysis through comparative testing of a complex-valued neural network designed to classify datasets containing real values. The proposed network will consist of an input layer, which will utilise a circular(sine) function to map the real-valued input onto the complex plane, followed by a hidden layer employing a Gaussian-like sech activation function, followed by the output layer consisting of a single neuron, with encoded outputs corresponding to various class label used to depict the classification of the input data. The training process will consist of the Least Mean Square Error minimization problem, with the error being sought to be minimized between the obtained output and the encoded desired outputs. It will be shown during the presentation of the testing results that the network design performs competitively with real-valued as well as complex-valued designs, and could provide a foundation for building improvements on the faster performing Circular Complex-Valued Neural Networks.