Sparse kernel orthonormalized PLS for feature extraction in large datasets

Jerónimo Arenas-García, Kaare Brandt Petersen, Lars Kai Hansen

    Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

    222 Downloads (Pure)

    Abstract

    In this paper we are presenting a novel multivariate analysis method for large scale problems. Our scheme is based on a novel kernel orthonormalized partial least squares (PLS) variant for feature extraction, imposing sparsity constrains in the solution to improve scalability. The algorithm is tested on a benchmark of UCI data sets, and on the analysis of integrated short-time music features for genre prediction. The upshot is that the method has strong expressive power even with rather few features, is clearly outperforming the ordinary kernel PLS, and therefore is an appealing method for feature extraction of labelled data.
    Original languageEnglish
    Title of host publicationNIPS 2006
    Publication date2006
    Publication statusPublished - 2006
    EventNIPS 2006 -
    Duration: 1 Jan 2006 → …

    Conference

    ConferenceNIPS 2006
    Period01/01/2006 → …

    Cite this

    Arenas-García, Jerónimo ; Petersen, Kaare Brandt ; Hansen, Lars Kai. / Sparse kernel orthonormalized PLS for feature extraction in large datasets. NIPS 2006. 2006.
    @inproceedings{f4ce1c919d3c42b68e4597d8deedb55c,
    title = "Sparse kernel orthonormalized PLS for feature extraction in large datasets",
    abstract = "In this paper we are presenting a novel multivariate analysis method for large scale problems. Our scheme is based on a novel kernel orthonormalized partial least squares (PLS) variant for feature extraction, imposing sparsity constrains in the solution to improve scalability. The algorithm is tested on a benchmark of UCI data sets, and on the analysis of integrated short-time music features for genre prediction. The upshot is that the method has strong expressive power even with rather few features, is clearly outperforming the ordinary kernel PLS, and therefore is an appealing method for feature extraction of labelled data.",
    author = "Jer{\'o}nimo Arenas-Garc{\'i}a and Petersen, {Kaare Brandt} and Hansen, {Lars Kai}",
    year = "2006",
    language = "English",
    booktitle = "NIPS 2006",

    }

    Arenas-García, J, Petersen, KB & Hansen, LK 2006, Sparse kernel orthonormalized PLS for feature extraction in large datasets. in NIPS 2006. NIPS 2006, 01/01/2006.

    Sparse kernel orthonormalized PLS for feature extraction in large datasets. / Arenas-García, Jerónimo; Petersen, Kaare Brandt; Hansen, Lars Kai.

    NIPS 2006. 2006.

    Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

    TY - GEN

    T1 - Sparse kernel orthonormalized PLS for feature extraction in large datasets

    AU - Arenas-García, Jerónimo

    AU - Petersen, Kaare Brandt

    AU - Hansen, Lars Kai

    PY - 2006

    Y1 - 2006

    N2 - In this paper we are presenting a novel multivariate analysis method for large scale problems. Our scheme is based on a novel kernel orthonormalized partial least squares (PLS) variant for feature extraction, imposing sparsity constrains in the solution to improve scalability. The algorithm is tested on a benchmark of UCI data sets, and on the analysis of integrated short-time music features for genre prediction. The upshot is that the method has strong expressive power even with rather few features, is clearly outperforming the ordinary kernel PLS, and therefore is an appealing method for feature extraction of labelled data.

    AB - In this paper we are presenting a novel multivariate analysis method for large scale problems. Our scheme is based on a novel kernel orthonormalized partial least squares (PLS) variant for feature extraction, imposing sparsity constrains in the solution to improve scalability. The algorithm is tested on a benchmark of UCI data sets, and on the analysis of integrated short-time music features for genre prediction. The upshot is that the method has strong expressive power even with rather few features, is clearly outperforming the ordinary kernel PLS, and therefore is an appealing method for feature extraction of labelled data.

    M3 - Article in proceedings

    BT - NIPS 2006

    ER -