### Abstract

The a-divergence is utilized to derive a generalized expectation and maximization algorithm (EM algorithm). This algorithm has a wide range of applications. In this paper, neural network learning for mixture probabilities is focused. The a-EM algorithm includes the existing EM algorithm as a special case since that corresponds to a = -1. The parameter a specifies a probability weight for the learning. This number affects learning speed and local optimality. In the discussions of update equations of neural nets, extensions of basic statistics such as Fisher's efficient score, his measure of information and Cramdr-Rao's inequality are also given. Besides, this paper unveils another new idea. It is found that the cyclic EM structure can be used as a building block to generate a learning systolic array. Attaching monitors to this systolic array makes it possible to create a functionally distributed learning system.

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 | 483-492 |

Number of pages | 10 |

Volume | 1240 LNCS |

ISBN (Print) | 3540630473, 9783540630470 |

Publication status | Published - 1997 |

Event | 4th International Work-Conference on Artificial and Natural Neural Networks, IWANN 1997 - Lanzarote, Canary Islands Duration: 1997 Jun 4 → 1997 Jun 6 |

### Publication series

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

Volume | 1240 LNCS |

ISSN (Print) | 03029743 |

ISSN (Electronic) | 16113349 |

### Other

Other | 4th International Work-Conference on Artificial and Natural Neural Networks, IWANN 1997 |
---|---|

City | Lanzarote, Canary Islands |

Period | 97/6/4 → 97/6/6 |

### 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. 1240 LNCS, pp. 483-492). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 1240 LNCS). Springer Verlag.

**The α-EM algorithm : A block connectable generalized leaning tool for neural networks.** / Matsuyama, Yasuo.

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. 1240 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 1240 LNCS, Springer Verlag, pp. 483-492, 4th International Work-Conference on Artificial and Natural Neural Networks, IWANN 1997, Lanzarote, Canary Islands, 97/6/4.

}

TY - GEN

T1 - The α-EM algorithm

T2 - A block connectable generalized leaning tool for neural networks

AU - Matsuyama, Yasuo

PY - 1997

Y1 - 1997

N2 - The a-divergence is utilized to derive a generalized expectation and maximization algorithm (EM algorithm). This algorithm has a wide range of applications. In this paper, neural network learning for mixture probabilities is focused. The a-EM algorithm includes the existing EM algorithm as a special case since that corresponds to a = -1. The parameter a specifies a probability weight for the learning. This number affects learning speed and local optimality. In the discussions of update equations of neural nets, extensions of basic statistics such as Fisher's efficient score, his measure of information and Cramdr-Rao's inequality are also given. Besides, this paper unveils another new idea. It is found that the cyclic EM structure can be used as a building block to generate a learning systolic array. Attaching monitors to this systolic array makes it possible to create a functionally distributed learning system.

AB - The a-divergence is utilized to derive a generalized expectation and maximization algorithm (EM algorithm). This algorithm has a wide range of applications. In this paper, neural network learning for mixture probabilities is focused. The a-EM algorithm includes the existing EM algorithm as a special case since that corresponds to a = -1. The parameter a specifies a probability weight for the learning. This number affects learning speed and local optimality. In the discussions of update equations of neural nets, extensions of basic statistics such as Fisher's efficient score, his measure of information and Cramdr-Rao's inequality are also given. Besides, this paper unveils another new idea. It is found that the cyclic EM structure can be used as a building block to generate a learning systolic array. Attaching monitors to this systolic array makes it possible to create a functionally distributed learning system.

UR - http://www.scopus.com/inward/record.url?scp=6744266982&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=6744266982&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:6744266982

SN - 3540630473

SN - 9783540630470

VL - 1240 LNCS

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

SP - 483

EP - 492

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

PB - Springer Verlag

ER -