Recurrent neural network training by nprKF joint estimation

被引：3

作者：

Feldkamp, LA ^{[1
]}

Feldkamp, TM ^{[1
]}

Prokhorov, DV ^{[1
]}

机构：

[1] Ford Motor Co, Ford Res Lab, Dearborn, MI 48121 USA

来源：

PROCEEDING OF THE 2002 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, VOLS 1-3 | 2002年

关键词：

D O I：

10.1109/IJCNN.2002.1007463

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

We present a method for training recurrent networks with the joint estimation of states and parameters, using the "derivative-free" formulation for nonlinear Kalman filters by Norgaard, Poulsen, and Ravn (NPR) [1]. Our approach is consistent with that described by Williams [2] for the extended Kalman filter (EKF). We have extended the treatment to handle multistream training and propose ways of making the required computation more efficient.

引用

页码：2086 / 2091

页数：2

共 8 条

[1] A signal processing framework based on dynamic neural networks with application to problems in adaptation, filtering, and classification [J].

Feldkamp, LA ;

Puskorius, GV .

PROCEEDINGS OF THE IEEE, 1998, 86 (11) :2259-2277

[2]

Feldkamp LA, 2001, IEEE IJCNN, P109, DOI 10.1109/IJCNN.2001.939001

[3]

MATTHEWS MB, 1990, P INT NEUR NETW C PA, V1, P115

[4]

NORGAARD M, 2000, IMMREP199815 TU DENM

[5]

Wan EA, 2001, ADAPT LEARN SYST SIG, P221

[6]

Wan EA, 2001, ADAPT LEARN SYST SIG, P123

[7] The unscented Kalman Filter for nonlinear estimation [J].

Wan, EA ;

van der Merwe, R .

IEEE 2000 ADAPTIVE SYSTEMS FOR SIGNAL PROCESSING, COMMUNICATIONS, AND CONTROL SYMPOSIUM - PROCEEDINGS, 2000, :153-158

[8]

WILLIAMS RJ, 1992, P INT JOINT C NEUR N, V4, P241

← 1 →