### Abstract

This paper discusses the problem of applying sigmoidal neural networks to identification of nonlinear dynamical systems. When using sigmoidal neural networks directly as nonlinear models, one often meets problems such as model parameters lack of physical meaning, sensitivity to noise in model training. In this paper, we introduce an embedded sigmoidal neural network model, in which the neural network is not used directly as a model, but is embedded in a shield such that part of the model parameters become meaningful. Corresponding to the meaningful part and the meaningless part of model parameters, a hierarchical training algorithm consisting of two learning loops is then introduced to train the model. Simulation results show that such a dual loop learning algorithm can solve the noise sensitivity and local minimum problems to some extent.

Original language | English |
---|---|

Title of host publication | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |

Publisher | Springer Verlag |

Pages | 937-942 |

Number of pages | 6 |

Volume | 2130 |

ISBN (Print) | 3540424865, 9783540446682 |

DOIs | |

Publication status | Published - 2001 |

Externally published | Yes |

Event | International Conference on Artificial Neural Networks, ICANN 2001 - Vienna, Austria Duration: 2001 Aug 21 → 2001 Aug 25 |

### Publication series

Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
---|---|

Volume | 2130 |

ISSN (Print) | 03029743 |

ISSN (Electronic) | 16113349 |

### Other

Other | International Conference on Artificial Neural Networks, ICANN 2001 |
---|---|

Country | Austria |

City | Vienna |

Period | 01/8/21 → 01/8/25 |

### Fingerprint

### ASJC Scopus subject areas

- Computer Science(all)
- Theoretical Computer Science

### Cite this

*Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)*(Vol. 2130, pp. 937-942). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2130). Springer Verlag. https://doi.org/10.1007/3-540-44668-0_129

**A hierarchical method for training embedded sigmoidal neural networks.** / Furuzuki, Takayuki; Hirasawa, Kotaro.

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

*Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics).*vol. 2130, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 2130, Springer Verlag, pp. 937-942, International Conference on Artificial Neural Networks, ICANN 2001, Vienna, Austria, 01/8/21. https://doi.org/10.1007/3-540-44668-0_129

}

TY - GEN

T1 - A hierarchical method for training embedded sigmoidal neural networks

AU - Furuzuki, Takayuki

AU - Hirasawa, Kotaro

PY - 2001

Y1 - 2001

N2 - This paper discusses the problem of applying sigmoidal neural networks to identification of nonlinear dynamical systems. When using sigmoidal neural networks directly as nonlinear models, one often meets problems such as model parameters lack of physical meaning, sensitivity to noise in model training. In this paper, we introduce an embedded sigmoidal neural network model, in which the neural network is not used directly as a model, but is embedded in a shield such that part of the model parameters become meaningful. Corresponding to the meaningful part and the meaningless part of model parameters, a hierarchical training algorithm consisting of two learning loops is then introduced to train the model. Simulation results show that such a dual loop learning algorithm can solve the noise sensitivity and local minimum problems to some extent.

AB - This paper discusses the problem of applying sigmoidal neural networks to identification of nonlinear dynamical systems. When using sigmoidal neural networks directly as nonlinear models, one often meets problems such as model parameters lack of physical meaning, sensitivity to noise in model training. In this paper, we introduce an embedded sigmoidal neural network model, in which the neural network is not used directly as a model, but is embedded in a shield such that part of the model parameters become meaningful. Corresponding to the meaningful part and the meaningless part of model parameters, a hierarchical training algorithm consisting of two learning loops is then introduced to train the model. Simulation results show that such a dual loop learning algorithm can solve the noise sensitivity and local minimum problems to some extent.

UR - http://www.scopus.com/inward/record.url?scp=33645276097&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33645276097&partnerID=8YFLogxK

U2 - 10.1007/3-540-44668-0_129

DO - 10.1007/3-540-44668-0_129

M3 - Conference contribution

SN - 3540424865

SN - 9783540446682

VL - 2130

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 937

EP - 942

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

PB - Springer Verlag

ER -