Title: Exploring real domain problems on the second generation neural network

Authors: Amit Gupta; Bipin Kumar Tripathi; Vivek Srivastava

Addresses: Department of Computer Science and Engineering, Dr. A.P.J. Abdul Kalam Technical University, Lucknow, India ' Department of Computer Science and Engineering, Harcourt Butler Technical University, Kanpur, India ' Department of Computer Science and Engineering, Rama University, Kanpur, India

Abstract: This paper presents a competitive performance of second generation neural network (CVNN) on the two dimensional space over first generation neural network (RVNN) on single dimensional space. The real datasets problems are selected for proposed research work. The second-generation neural network is based on the theory of complex number. Complex numbers are forms of subset of real numbers having magnitude and phase to represent a real valued phenomenon. For the testing and training of real valued problems in complex domain, a mathematical approach Hilbert transformation is used to convert all the real valued data in complex form by sifting the phase by ±90 degree with same amplitude. For learning the network RBP and CBP algorithm is used over proposed benchmark datasets (both real and complex) to train the neural network. A new complex activation function (amplitude and phase type) is utilising by second generation neural network (CVNN). The results show improved efficacy and minimum number of learning cycles for Second generation neural network in complex domain over the first generation neural network in real domain.

Keywords: real value neural network; complex value neural network; complex activation function; back propagation algorithm; Hilbert transformation.

DOI: 10.1504/IJAIP.2023.135028

International Journal of Advanced Intelligence Paradigms, 2023 Vol.26 No.2, pp.143 - 157

Received: 08 Mar 2018
Accepted: 04 Apr 2018
Published online: 28 Nov 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article