Information geometric perspective of modal linear regression

Keishi Sando, Shotaro Akaho, Noboru Murata, Hideitsu Hino

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

    Abstract

    Modal linear regression (MLR) is a standard method for modeling the conditional mode of a response variable using a linear combination of explanatory variables. It is effective when dealing with response variables with an asymmetric, multi-modal distribution. Because of the nonparametric nature of MLR, it is difficult to construct a statistical model manifold in the sense of information geometry. In this work, a model manifold is constructed using observations instead of explicit parametric models. We also propose a method for constructing a data manifold based on an empirical distribution. The em algorithm, which is a geometric formulation of the EM algorithm, of MLR is shown to be equivalent to the conventional EM algorithm of MLR.

    Original languageEnglish
    Title of host publicationNeural Information Processing - 25th International Conference, ICONIP 2018, Proceedings
    EditorsLong Cheng, Seiichi Ozawa, Andrew Chi Sing Leung
    PublisherSpringer-Verlag
    Pages535-545
    Number of pages11
    ISBN (Print)9783030041816
    DOIs
    Publication statusPublished - 2018 Jan 1
    Event25th International Conference on Neural Information Processing, ICONIP 2018 - Siem Reap, Cambodia
    Duration: 2018 Dec 132018 Dec 16

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume11303 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Other

    Other25th International Conference on Neural Information Processing, ICONIP 2018
    CountryCambodia
    CitySiem Reap
    Period18/12/1318/12/16

    Fingerprint

    Linear regression
    EM Algorithm
    Information Geometry
    Empirical Distribution
    Parametric Model
    Statistical Model
    Linear Combination
    Geometry
    Formulation
    Modeling
    Model

    Keywords

    • EM algorithm
    • Information geometry
    • Modal linear regression

    ASJC Scopus subject areas

    • Theoretical Computer Science
    • Computer Science(all)

    Cite this

    Sando, K., Akaho, S., Murata, N., & Hino, H. (2018). Information geometric perspective of modal linear regression. In L. Cheng, S. Ozawa, & A. C. S. Leung (Eds.), Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings (pp. 535-545). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11303 LNCS). Springer-Verlag. https://doi.org/10.1007/978-3-030-04182-3_47

    Information geometric perspective of modal linear regression. / Sando, Keishi; Akaho, Shotaro; Murata, Noboru; Hino, Hideitsu.

    Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. ed. / Long Cheng; Seiichi Ozawa; Andrew Chi Sing Leung. Springer-Verlag, 2018. p. 535-545 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11303 LNCS).

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

    Sando, K, Akaho, S, Murata, N & Hino, H 2018, Information geometric perspective of modal linear regression. in L Cheng, S Ozawa & ACS Leung (eds), Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11303 LNCS, Springer-Verlag, pp. 535-545, 25th International Conference on Neural Information Processing, ICONIP 2018, Siem Reap, Cambodia, 18/12/13. https://doi.org/10.1007/978-3-030-04182-3_47
    Sando K, Akaho S, Murata N, Hino H. Information geometric perspective of modal linear regression. In Cheng L, Ozawa S, Leung ACS, editors, Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. Springer-Verlag. 2018. p. 535-545. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-04182-3_47
    Sando, Keishi ; Akaho, Shotaro ; Murata, Noboru ; Hino, Hideitsu. / Information geometric perspective of modal linear regression. Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. editor / Long Cheng ; Seiichi Ozawa ; Andrew Chi Sing Leung. Springer-Verlag, 2018. pp. 535-545 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
    @inproceedings{2cb8d67e997e4a379bf1189eb9beab7f,
    title = "Information geometric perspective of modal linear regression",
    abstract = "Modal linear regression (MLR) is a standard method for modeling the conditional mode of a response variable using a linear combination of explanatory variables. It is effective when dealing with response variables with an asymmetric, multi-modal distribution. Because of the nonparametric nature of MLR, it is difficult to construct a statistical model manifold in the sense of information geometry. In this work, a model manifold is constructed using observations instead of explicit parametric models. We also propose a method for constructing a data manifold based on an empirical distribution. The em algorithm, which is a geometric formulation of the EM algorithm, of MLR is shown to be equivalent to the conventional EM algorithm of MLR.",
    keywords = "EM algorithm, Information geometry, Modal linear regression",
    author = "Keishi Sando and Shotaro Akaho and Noboru Murata and Hideitsu Hino",
    year = "2018",
    month = "1",
    day = "1",
    doi = "10.1007/978-3-030-04182-3_47",
    language = "English",
    isbn = "9783030041816",
    series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
    publisher = "Springer-Verlag",
    pages = "535--545",
    editor = "Long Cheng and Seiichi Ozawa and Leung, {Andrew Chi Sing}",
    booktitle = "Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings",

    }

    TY - GEN

    T1 - Information geometric perspective of modal linear regression

    AU - Sando, Keishi

    AU - Akaho, Shotaro

    AU - Murata, Noboru

    AU - Hino, Hideitsu

    PY - 2018/1/1

    Y1 - 2018/1/1

    N2 - Modal linear regression (MLR) is a standard method for modeling the conditional mode of a response variable using a linear combination of explanatory variables. It is effective when dealing with response variables with an asymmetric, multi-modal distribution. Because of the nonparametric nature of MLR, it is difficult to construct a statistical model manifold in the sense of information geometry. In this work, a model manifold is constructed using observations instead of explicit parametric models. We also propose a method for constructing a data manifold based on an empirical distribution. The em algorithm, which is a geometric formulation of the EM algorithm, of MLR is shown to be equivalent to the conventional EM algorithm of MLR.

    AB - Modal linear regression (MLR) is a standard method for modeling the conditional mode of a response variable using a linear combination of explanatory variables. It is effective when dealing with response variables with an asymmetric, multi-modal distribution. Because of the nonparametric nature of MLR, it is difficult to construct a statistical model manifold in the sense of information geometry. In this work, a model manifold is constructed using observations instead of explicit parametric models. We also propose a method for constructing a data manifold based on an empirical distribution. The em algorithm, which is a geometric formulation of the EM algorithm, of MLR is shown to be equivalent to the conventional EM algorithm of MLR.

    KW - EM algorithm

    KW - Information geometry

    KW - Modal linear regression

    UR - http://www.scopus.com/inward/record.url?scp=85059008645&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85059008645&partnerID=8YFLogxK

    U2 - 10.1007/978-3-030-04182-3_47

    DO - 10.1007/978-3-030-04182-3_47

    M3 - Conference contribution

    AN - SCOPUS:85059008645

    SN - 9783030041816

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

    SP - 535

    EP - 545

    BT - Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings

    A2 - Cheng, Long

    A2 - Ozawa, Seiichi

    A2 - Leung, Andrew Chi Sing

    PB - Springer-Verlag

    ER -