Improving the Approximation Smoothness of Radial Basis Neural Networks
Anthony Little and Leonid Reznik
School of Communications & Informatics, Victoria University of Technology, P.O. Box 14428, Melbourne City MC, VIC 8001 Australia
Implementation of industrial multimedia and control systems, which can be described as non-linear two-dimensional surfaces, may be expensive in terms of both hardware and software requirements. Radial Basis neural networks are excellent platforms for implementing approximations of such surfaces. This paper proposes an alternative basis function in an effort to improve the ability of the network to provide a smooth surface. The resulting basis function presents a balanced approach to the integration between hidden layer neurons, while maintaining a computational structure suitable for low cost software implementation. The paper investigates the proposed structure in regard to different criteria
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 Internationa License.