LEARNING ALGORITHMS OF LAYERED NEURAL NETWORKS VIA EXTENDED KALMAN FILTERS

WATANABE, K; FUKUDA, T; TZAFESTAS, SG

dc.contributor.author	WATANABE, K	en
dc.contributor.author	FUKUDA, T	en
dc.contributor.author	TZAFESTAS, SG	en
dc.date.accessioned	2014-03-01T01:08:25Z
dc.date.available	2014-03-01T01:08:25Z
dc.date.issued	1991	en
dc.identifier.issn	0020-7721	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/10475
dc.subject	extended kalman filter	en
dc.subject	Learning Algorithm	en
dc.subject	Neural Network	en
dc.subject.classification	Automation & Control Systems	en
dc.subject.classification	Computer Science, Theory & Methods	en
dc.subject.classification	Operations Research & Management Science	en
dc.title	LEARNING ALGORITHMS OF LAYERED NEURAL NETWORKS VIA EXTENDED KALMAN FILTERS	en
heal.type	journalArticle	en
heal.identifier.primary	10.1080/00207729108910654	en
heal.identifier.secondary	http://dx.doi.org/10.1080/00207729108910654	en
heal.language	English	en
heal.publicationDate	1991	en
heal.abstract	Learning algorithms are described for layered feedforward type neural networks, in which a unit generates a real-valued output through a logistic function. The problem of adjusting the weights of internal hidden units can be regarded as a problem of estimating (or identifying) constant parametes with a non-linear observation equation. The present algorithm based on the extended Kalman filter has just the time-varying learning rate, while the well-known back-propagation (or generalized delta rule) algorithm based on gradient descent has a constant learning rate. From some simulation examples it is shown that when a sufficiently trained network is desired, the learning speed of the proposed algorithm is faster than that of the traditional back-propagation algorithm.	en
heal.publisher	TAYLOR & FRANCIS LTD	en
heal.journalName	INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE	en
dc.identifier.doi	10.1080/00207729108910654	en
dc.identifier.isi	ISI:A1991FC61500012	en
dc.identifier.volume	22	en
dc.identifier.issue	4	en
dc.identifier.spage	753	en
dc.identifier.epage	768	en