Accuracy of surrogate solutions of integral equations by feedforward networks

Authors: Věra Kůrková
Publication year: 2014
Publication type: Kapitola v monografii
Periodical: Issues and Challenges of Intelligent Systems and Computational Intelligence
Projects: Learning of functional relationships from high-dimensional data
Modeling of complex systems for softcomputing methods
Bibliographic citation: KŮRKOVÁ, Věra. Accuracy of surrogate solutions of integral equations by feedforward networks. In: KÓCZY, L., POZNA, C.R., KACPRZYK, J., eds. Issues and Challenges of Intelligent Systems and Computational Intelligence. Cham: Springer, 2014, S. 91-102. Studies in Computational Intelligence, 530. ISBN 978-3-319-03205-4
Dostupné z: doi:10.1007/978-3-319-03206-1_7
DOI: 10.1007/978-3-319-03206-1_7
ID: 398349

Abstract

Surrogate solutions of Fredholm integral equations by feedforward neural networks are investigated theoretically. Convergence of surrogate solutions computable by networks with increasing numbers of computational units to theoretically optimal solutions is proven and upper bounds on rates of convergence are derived. The results hold for a variety of computational units, they are illustrated by examples of perceptrons and Gaussian radial units.