α-EM algorithm and its basic properties

Yasuo Matsuyama

    Research output: Contribution to journalArticle

    Abstract

    A new class of statistical algorithms is presented and examined. The method is called the α-EM algorithm. This novel algorithm contains the traditional EM algorithm as a special case of α = -1. The choice of the design parameter `α' affects the eigenvalues of Hessian matrices for likelihood maximization. This causes much faster convergence than the traditional EM algorithm. Convergence theorems are given for the basic α-EM algorithm and its practical variants. Numerical evaluation shows fast convergence at nearly one-third the iteration counts and one-half the CPU time relative to the traditional method.

    Original languageEnglish
    Pages (from-to)12-23
    Number of pages12
    JournalSystems and Computers in Japan
    Volume31
    Issue number11
    DOIs
    Publication statusPublished - 2000 Oct

    Fingerprint

    EM Algorithm
    Hessian matrix
    CPU Time
    Parameter Design
    Convergence Theorem
    Likelihood
    Count
    Eigenvalue
    Iteration
    Program processors
    Evaluation

    ASJC Scopus subject areas

    • Computational Theory and Mathematics
    • Hardware and Architecture
    • Information Systems
    • Theoretical Computer Science

    Cite this

    α-EM algorithm and its basic properties. / Matsuyama, Yasuo.

    In: Systems and Computers in Japan, Vol. 31, No. 11, 10.2000, p. 12-23.

    Research output: Contribution to journalArticle

    Matsuyama, Yasuo. / α-EM algorithm and its basic properties. In: Systems and Computers in Japan. 2000 ; Vol. 31, No. 11. pp. 12-23.
    @article{74d94394c7e2438abb3b97985615752b,
    title = "α-EM algorithm and its basic properties",
    abstract = "A new class of statistical algorithms is presented and examined. The method is called the α-EM algorithm. This novel algorithm contains the traditional EM algorithm as a special case of α = -1. The choice of the design parameter `α' affects the eigenvalues of Hessian matrices for likelihood maximization. This causes much faster convergence than the traditional EM algorithm. Convergence theorems are given for the basic α-EM algorithm and its practical variants. Numerical evaluation shows fast convergence at nearly one-third the iteration counts and one-half the CPU time relative to the traditional method.",
    author = "Yasuo Matsuyama",
    year = "2000",
    month = "10",
    doi = "10.1002/1520-684X(200010)31:11<12::AID-SCJ2>3.0.CO;2-O",
    language = "English",
    volume = "31",
    pages = "12--23",
    journal = "Systems and Computers in Japan",
    issn = "0882-1666",
    publisher = "John Wiley and Sons Inc.",
    number = "11",

    }

    TY - JOUR

    T1 - α-EM algorithm and its basic properties

    AU - Matsuyama, Yasuo

    PY - 2000/10

    Y1 - 2000/10

    N2 - A new class of statistical algorithms is presented and examined. The method is called the α-EM algorithm. This novel algorithm contains the traditional EM algorithm as a special case of α = -1. The choice of the design parameter `α' affects the eigenvalues of Hessian matrices for likelihood maximization. This causes much faster convergence than the traditional EM algorithm. Convergence theorems are given for the basic α-EM algorithm and its practical variants. Numerical evaluation shows fast convergence at nearly one-third the iteration counts and one-half the CPU time relative to the traditional method.

    AB - A new class of statistical algorithms is presented and examined. The method is called the α-EM algorithm. This novel algorithm contains the traditional EM algorithm as a special case of α = -1. The choice of the design parameter `α' affects the eigenvalues of Hessian matrices for likelihood maximization. This causes much faster convergence than the traditional EM algorithm. Convergence theorems are given for the basic α-EM algorithm and its practical variants. Numerical evaluation shows fast convergence at nearly one-third the iteration counts and one-half the CPU time relative to the traditional method.

    UR - http://www.scopus.com/inward/record.url?scp=0034297611&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=0034297611&partnerID=8YFLogxK

    U2 - 10.1002/1520-684X(200010)31:11<12::AID-SCJ2>3.0.CO;2-O

    DO - 10.1002/1520-684X(200010)31:11<12::AID-SCJ2>3.0.CO;2-O

    M3 - Article

    VL - 31

    SP - 12

    EP - 23

    JO - Systems and Computers in Japan

    JF - Systems and Computers in Japan

    SN - 0882-1666

    IS - 11

    ER -