Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter

Yingxin Liao; Min Wu; Jinhua She; Kaoru Hirota

doi:10.1109/CHICC.2008.4605166

Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter

Yingxin Liao^*
, Min Wu
, Jinhua She
, Kaoru Hirota

^*Corresponding author for this work

Central South University of Forestry & Technology
Central South University
Tokyo University of Technology
Tokyo Institute of Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

3 Citations (Scopus)

Abstract

This paper presents a sequential growing-and-pruning learning algorithm employing an unscented or extended Kalman filter (SGAPL-UKF or SGAPL-EKF) for a recurrent neural network (RNN). The RNN is constructed using a sequential-learning algorithm that employs growing-and-pruning (GAP) criteria based on the concept of the significance of hidden neurons to yield a compact network; and an unscented or extended Kalman filter improves the learning accuracy by providing estimates of the parameters of the RNN from incomplete samples. As an example, this method was used to estimate the output of a Mackey-Glass time series. A comparison of the results obtained with a UKF and an EKF yielded guidelines about which situations each type of filter is suitable for. Verification results show the effectiveness of the learning algorithm.

Original language	English
Title of host publication	Proceedings of the 27th Chinese Control Conference, CCC
Pages	242-247
Number of pages	6
DOIs	https://doi.org/10.1109/CHICC.2008.4605166
Publication status	Published - 2008
Externally published	Yes
Event	27th Chinese Control Conference, CCC - Kunming, Yunnan, China Duration: 16 Jul 2008 → 18 Jul 2008

Publication series

Name	Proceedings of the 27th Chinese Control Conference, CCC

Conference

Conference	27th Chinese Control Conference, CCC
Country/Territory	China
City	Kunming, Yunnan
Period	16/07/08 → 18/07/08

Keywords

Extended Kalman filter
Function approximation
Mackey-glass series
Recurrent neural network
Sequential learning
Unseented Kalman filter

Access to Document

10.1109/CHICC.2008.4605166

Cite this

@inproceedings{aa70c048ecbb4b219f0774f5f7fe93b0,

title = "Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter",

abstract = "This paper presents a sequential growing-and-pruning learning algorithm employing an unscented or extended Kalman filter (SGAPL-UKF or SGAPL-EKF) for a recurrent neural network (RNN). The RNN is constructed using a sequential-learning algorithm that employs growing-and-pruning (GAP) criteria based on the concept of the significance of hidden neurons to yield a compact network; and an unscented or extended Kalman filter improves the learning accuracy by providing estimates of the parameters of the RNN from incomplete samples. As an example, this method was used to estimate the output of a Mackey-Glass time series. A comparison of the results obtained with a UKF and an EKF yielded guidelines about which situations each type of filter is suitable for. Verification results show the effectiveness of the learning algorithm.",

keywords = "Extended Kalman filter, Function approximation, Mackey-glass series, Recurrent neural network, Sequential learning, Unseented Kalman filter",

author = "Yingxin Liao and Min Wu and Jinhua She and Kaoru Hirota",

year = "2008",

doi = "10.1109/CHICC.2008.4605166",

language = "English",

isbn = "9787900719706",

series = "Proceedings of the 27th Chinese Control Conference, CCC",

pages = "242--247",

booktitle = "Proceedings of the 27th Chinese Control Conference, CCC",

note = "27th Chinese Control Conference, CCC ; Conference date: 16-07-2008 Through 18-07-2008",

}

Liao, Y, Wu, M, She, J & Hirota, K 2008, Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter. in Proceedings of the 27th Chinese Control Conference, CCC., 4605166, Proceedings of the 27th Chinese Control Conference, CCC, pp. 242-247, 27th Chinese Control Conference, CCC, Kunming, Yunnan, China, 16/07/08. https://doi.org/10.1109/CHICC.2008.4605166

Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter. / Liao, Yingxin; Wu, Min; She, Jinhua et al.
Proceedings of the 27th Chinese Control Conference, CCC. 2008. p. 242-247 4605166 (Proceedings of the 27th Chinese Control Conference, CCC).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter

AU - Liao, Yingxin

AU - Wu, Min

AU - She, Jinhua

AU - Hirota, Kaoru

PY - 2008

Y1 - 2008

N2 - This paper presents a sequential growing-and-pruning learning algorithm employing an unscented or extended Kalman filter (SGAPL-UKF or SGAPL-EKF) for a recurrent neural network (RNN). The RNN is constructed using a sequential-learning algorithm that employs growing-and-pruning (GAP) criteria based on the concept of the significance of hidden neurons to yield a compact network; and an unscented or extended Kalman filter improves the learning accuracy by providing estimates of the parameters of the RNN from incomplete samples. As an example, this method was used to estimate the output of a Mackey-Glass time series. A comparison of the results obtained with a UKF and an EKF yielded guidelines about which situations each type of filter is suitable for. Verification results show the effectiveness of the learning algorithm.

AB - This paper presents a sequential growing-and-pruning learning algorithm employing an unscented or extended Kalman filter (SGAPL-UKF or SGAPL-EKF) for a recurrent neural network (RNN). The RNN is constructed using a sequential-learning algorithm that employs growing-and-pruning (GAP) criteria based on the concept of the significance of hidden neurons to yield a compact network; and an unscented or extended Kalman filter improves the learning accuracy by providing estimates of the parameters of the RNN from incomplete samples. As an example, this method was used to estimate the output of a Mackey-Glass time series. A comparison of the results obtained with a UKF and an EKF yielded guidelines about which situations each type of filter is suitable for. Verification results show the effectiveness of the learning algorithm.

KW - Extended Kalman filter

KW - Function approximation

KW - Mackey-glass series

KW - Recurrent neural network

KW - Sequential learning

KW - Unseented Kalman filter

UR - https://www.scopus.com/pages/publications/52449126625

U2 - 10.1109/CHICC.2008.4605166

DO - 10.1109/CHICC.2008.4605166

M3 - Conference contribution

AN - SCOPUS:52449126625

SN - 9787900719706

T3 - Proceedings of the 27th Chinese Control Conference, CCC

SP - 242

EP - 247

BT - Proceedings of the 27th Chinese Control Conference, CCC

T2 - 27th Chinese Control Conference, CCC

Y2 - 16 July 2008 through 18 July 2008

ER -

Sequential growing-and-pruning learning for recurrent neural networks using unseented or extended Kalman filter

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this