Home // ICSEA 2014, The Ninth International Conference on Software Engineering Advances // View article
Authors:
Mikael Fridenfalk
Keywords: analytic; FNN; large-scale; least square method; neural network; robust; sigmoid
Abstract:
The multilayer feedforward neural network is presently one of the most popular computational methods in computer science. The current method for the evaluation of its weights is however performed by a relatively slow iterative method known as backpropagation. According to previous research, attempts to evaluate the weights analytically by the linear least square method, showed to accelerate the evaluation process significantly. The evaluated networks showed however to fail in robustness tests compared to well-trained networks by backpropagation, thus resembling overtrained networks. This paper presents the design and verification of a new method, that solves the robustness issues for a large-scale neural network with many hidden nodes, as an upgrade to the previously suggested analytic method.
Pages: 374 to 378
Copyright: Copyright (c) IARIA, 2014
Publication date: October 12, 2014
Published in: conference
ISSN: 2308-4235
ISBN: 978-1-61208-367-4
Location: Nice, France
Dates: from October 12, 2014 to October 16, 2014