Improvement of training method for dynamic neural network

Kunihiko Nakazono, Kouhei Ohnishi, Hiroshi Kinjo, Tetsuhiko Yamamoto

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this research, we propose a dynamic neural network (DNN) with the characteristics of stiffness, viscosity, and inertia and a training algorithm based on the back-propagation (BP) method to include a momentum term. In a previous research, we proposed a training algorithm for the DNN based on the BP method or GA-based training method. However, in the previous method it was necessary to determine the values of the DNN parameters by trial and error. So, the modified BP method and GA-based training method were designed to train not only the connecting weights but also the property parameters of the DNN. We develop the BP method to include a momentum term in order to increase the convergence of the training effect. Simulation results show that the DNN with characteristics of stiffness, viscosity, and inertia trained by the modified BP method to include the momentum term obtains good training performances for time series signals generated from periodic function. In this paper, we compare the DNN with a conventional training method in order to verify the effectiveness of the DNN.

Original languageEnglish
Title of host publicationProceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07
Pages324-327
Number of pages4
Publication statusPublished - 2007
Event12th International Symposium on Artificial Life and Robotics, AROB 12th'07 - Oita, Japan
Duration: 2007 Jan 252007 Jan 27

Other

Other12th International Symposium on Artificial Life and Robotics, AROB 12th'07
CountryJapan
CityOita
Period07/1/2507/1/27

Fingerprint

Backpropagation
Neural networks
Momentum
Stiffness
Viscosity
Time series

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Human-Computer Interaction

Cite this

Nakazono, K., Ohnishi, K., Kinjo, H., & Yamamoto, T. (2007). Improvement of training method for dynamic neural network. In Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07 (pp. 324-327)

Improvement of training method for dynamic neural network. / Nakazono, Kunihiko; Ohnishi, Kouhei; Kinjo, Hiroshi; Yamamoto, Tetsuhiko.

Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07. 2007. p. 324-327.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Nakazono, K, Ohnishi, K, Kinjo, H & Yamamoto, T 2007, Improvement of training method for dynamic neural network. in Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07. pp. 324-327, 12th International Symposium on Artificial Life and Robotics, AROB 12th'07, Oita, Japan, 07/1/25.
Nakazono K, Ohnishi K, Kinjo H, Yamamoto T. Improvement of training method for dynamic neural network. In Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07. 2007. p. 324-327
Nakazono, Kunihiko ; Ohnishi, Kouhei ; Kinjo, Hiroshi ; Yamamoto, Tetsuhiko. / Improvement of training method for dynamic neural network. Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07. 2007. pp. 324-327
@inproceedings{c00839e59a104e5982fbb4030c679206,
title = "Improvement of training method for dynamic neural network",
abstract = "In this research, we propose a dynamic neural network (DNN) with the characteristics of stiffness, viscosity, and inertia and a training algorithm based on the back-propagation (BP) method to include a momentum term. In a previous research, we proposed a training algorithm for the DNN based on the BP method or GA-based training method. However, in the previous method it was necessary to determine the values of the DNN parameters by trial and error. So, the modified BP method and GA-based training method were designed to train not only the connecting weights but also the property parameters of the DNN. We develop the BP method to include a momentum term in order to increase the convergence of the training effect. Simulation results show that the DNN with characteristics of stiffness, viscosity, and inertia trained by the modified BP method to include the momentum term obtains good training performances for time series signals generated from periodic function. In this paper, we compare the DNN with a conventional training method in order to verify the effectiveness of the DNN.",
author = "Kunihiko Nakazono and Kouhei Ohnishi and Hiroshi Kinjo and Tetsuhiko Yamamoto",
year = "2007",
language = "English",
isbn = "9784990288013",
pages = "324--327",
booktitle = "Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07",

}

TY - GEN

T1 - Improvement of training method for dynamic neural network

AU - Nakazono, Kunihiko

AU - Ohnishi, Kouhei

AU - Kinjo, Hiroshi

AU - Yamamoto, Tetsuhiko

PY - 2007

Y1 - 2007

N2 - In this research, we propose a dynamic neural network (DNN) with the characteristics of stiffness, viscosity, and inertia and a training algorithm based on the back-propagation (BP) method to include a momentum term. In a previous research, we proposed a training algorithm for the DNN based on the BP method or GA-based training method. However, in the previous method it was necessary to determine the values of the DNN parameters by trial and error. So, the modified BP method and GA-based training method were designed to train not only the connecting weights but also the property parameters of the DNN. We develop the BP method to include a momentum term in order to increase the convergence of the training effect. Simulation results show that the DNN with characteristics of stiffness, viscosity, and inertia trained by the modified BP method to include the momentum term obtains good training performances for time series signals generated from periodic function. In this paper, we compare the DNN with a conventional training method in order to verify the effectiveness of the DNN.

AB - In this research, we propose a dynamic neural network (DNN) with the characteristics of stiffness, viscosity, and inertia and a training algorithm based on the back-propagation (BP) method to include a momentum term. In a previous research, we proposed a training algorithm for the DNN based on the BP method or GA-based training method. However, in the previous method it was necessary to determine the values of the DNN parameters by trial and error. So, the modified BP method and GA-based training method were designed to train not only the connecting weights but also the property parameters of the DNN. We develop the BP method to include a momentum term in order to increase the convergence of the training effect. Simulation results show that the DNN with characteristics of stiffness, viscosity, and inertia trained by the modified BP method to include the momentum term obtains good training performances for time series signals generated from periodic function. In this paper, we compare the DNN with a conventional training method in order to verify the effectiveness of the DNN.

UR - http://www.scopus.com/inward/record.url?scp=78549259377&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=78549259377&partnerID=8YFLogxK

M3 - Conference contribution

SN - 9784990288013

SP - 324

EP - 327

BT - Proceedings of the 12th International Symposium on Artificial Life and Robotis, AROB 12th'07

ER -