The α-EM algorithm: A block connectable generalized leaning tool for neural networks

Yasuo Matsuyama

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

    7 Citations (Scopus)

    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 languageEnglish
    Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    PublisherSpringer Verlag
    Pages483-492
    Number of pages10
    Volume1240 LNCS
    ISBN (Print)3540630473, 9783540630470
    Publication statusPublished - 1997
    Event4th International Work-Conference on Artificial and Natural Neural Networks, IWANN 1997 - Lanzarote, Canary Islands
    Duration: 1997 Jun 41997 Jun 6

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume1240 LNCS
    ISSN (Print)03029743
    ISSN (Electronic)16113349

    Other

    Other4th International Work-Conference on Artificial and Natural Neural Networks, IWANN 1997
    CityLanzarote, Canary Islands
    Period97/6/497/6/6

    Fingerprint

    Neural Networks
    Neural networks
    Systolic Array
    Systolic arrays
    Local Optimality
    Measures of Information
    Neural Nets
    Learning Systems
    Building Blocks
    Learning systems
    Distributed Systems
    Divergence
    Monitor
    Update
    Statistics
    Learning
    Range of data

    ASJC Scopus subject areas

    • Computer Science(all)
    • Theoretical Computer Science

    Cite this

    Matsuyama, Y. (1997). The α-EM algorithm: A block connectable generalized leaning tool for neural networks. In 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.

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 1240 LNCS Springer Verlag, 1997. p. 483-492 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 1240 LNCS).

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

    Matsuyama, Y 1997, The α-EM algorithm: A block connectable generalized leaning tool for neural networks. in 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.
    Matsuyama Y. The α-EM algorithm: A block connectable generalized leaning tool for neural networks. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 1240 LNCS. Springer Verlag. 1997. p. 483-492. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
    Matsuyama, Yasuo. / The α-EM algorithm : A block connectable generalized leaning tool for neural networks. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 1240 LNCS Springer Verlag, 1997. pp. 483-492 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
    @inproceedings{d086c55600114cb88db527767635a3f7,
    title = "The α-EM algorithm: A block connectable generalized leaning tool for neural networks",
    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.",
    author = "Yasuo Matsuyama",
    year = "1997",
    language = "English",
    isbn = "3540630473",
    volume = "1240 LNCS",
    series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
    publisher = "Springer Verlag",
    pages = "483--492",
    booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

    }

    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 -