Full Length Research Paper

New activation functions for complex-valued neural network

Hamid A. Jalab¹* and Rabha W. Ibrahim²

¹Department of Computer System and Technology, Faculty of Computer Science and Information Technology, University Malaya, 50603 Kuala Lumpur, Malaysia.

²School of Mathematical Sciences, Faculty of Science and Technology, Universiti Kebangsaan Malaysia, Bangi 43600, Selangor Darul Ehsan, Malaysia.

*Corresponding author. E-mail: hamidjalab@um.edu.my

Accepted 14 March, 2011.

Abstract

This paper presents a new types of complex-valued sigmoid function for a fully multi-layered complex-valued neural network (CVNN). By using the concept of the subordination between analytic functions in open disc, we able to study the reducibility of CVNN. A real-world problem example has been used as a classifier. The simulations results reveal that the proposed fully complex-valued network, been better trained reduces the testing time by 54% compared to the choice of using the traditional sigmoid activation function.

Key words: Complex valued neural network, activation functions, reducibility, irreducibility, subordination.