Reverse neuron level decomposition for cooperative neuro - evolution of feedforward networks for time series prediction

Nand, Ravneil and Chandra, Rohitash (2016) Reverse neuron level decomposition for cooperative neuro - evolution of feedforward networks for time series prediction. In: Artificial Life and Computational Intelligence. Lecture Notes in Computer Science, 9592 . Springer International Publishing, Switzerland, pp. 171-182. ISBN 9783319282695

PDF - Submitted Version Restricted to Repository staff only Download (360kB)

Abstract

A major challenge in cooperative neuro-evolution is to find an efficient problem decomposition that takes into account architectural properties of the neural network and the training problem. In the past, neuron and synapse Level decomposition methods have shown promising results for time series problems, howsoever, the search for the optimal method remains. In this paper, a problem decomposition method, that is based on neuron level decomposition is proposed that features a reverse encoding scheme. It is used for training feedforward networks for time series prediction. The results show that the proposed method has improved performance when compared to related problem decomposition methods and shows competitive results when compared to related methods in the literature.

Item Type:	Book Chapter
Additional Information:	DOI: 10.1007/978-3-319-28270-1_15
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions:	Faculty of Science, Technology and Environment (FSTE) > School of Computing, Information and Mathematical Sciences
Depositing User:	Ravneil Nand
Date Deposited:	23 Aug 2016 00:12
Last Modified:	22 Mar 2017 02:51
URI:	https://repository.usp.ac.fj/id/eprint/9185

Actions (login required)

View Item