Presentation 1994/10/13
Learning Algorithm for Elman Networks based on Quasi-Newton Method
Kazumi Saito, Ryohei Nakano,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The BPTT and RTRL algorithms are representative learning methods for recurrent neural networks.Since these methods,however,are based on the steepest-descent algorithm,they often require a large number of iterations for convergence.In this paper,we propose a new learning algorithm called BPTTQ,which employs an efficient calculation of the optimal step lengths as the minimal points of an approximation.Experiments showed that BPTTQ worked much better than the existing algorithms.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Elman networks / BPTT / RTRL / quasi-Newton method / optimal step-length
Paper # NC94-38
Date of Issue

Conference Information
Committee NC
Conference Date 1994/10/13(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Vice Chair

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Learning Algorithm for Elman Networks based on Quasi-Newton Method
Sub Title (in English)
Keyword(1) Elman networks
Keyword(2) BPTT
Keyword(3) RTRL
Keyword(4) quasi-Newton method
Keyword(5) optimal step-length
1st Author's Name Kazumi Saito
1st Author's Affiliation Science Laboratories,NTT Communication()
2nd Author's Name Ryohei Nakano
2nd Author's Affiliation Science Laboratories,NTT Communication
Date 1994/10/13
Paper # NC94-38
Volume (vol) vol.94
Number (no) 272
Page pp.pp.-
#Pages 8
Date of Issue