Improved variance estimation along sample eigenvectors

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

    13 Downloads (Pure)

    Abstract

    Second order statistics estimates in the form of sample eigenvalues and sample eigenvectors give a sub optimal description of the population density. So far only attempts have been made to reduce the bias in the sample eigenvalues. However, because the sample eigenvectors differ from the population eigenvectors as well, the population eigenvalues are biased estimates of the variances along the sample eigenvectors. Therefore correction of the sample eigenvalues towards the population eigenvalues is not sufficient. The experiments in this paper show that replacing the sample eigenvalues with the variances along the sample eigenvectors often results in better estimates of the population density than replacing the sample eigenvalues with the population eigenvalues.
    Original languageUndefined
    Title of host publicationProceedings of the 30th Symposium on Information Theory in the Benelux
    Place of PublicationEindhoven
    PublisherWerkgemeenschap voor Informatie- en Communicatietheorie (WIC)
    Pages25-32
    Number of pages8
    ISBN (Print)978-90-386-1852-4
    Publication statusPublished - May 2009
    Event30th WIC Symposium on Information Theory in the Benelux 2009 - Eindhoven, Netherlands
    Duration: 28 May 200929 May 2009
    Conference number: 30

    Publication series

    Name
    PublisherWerkgemeenschap voor Informatie- en Communicatietheorie

    Conference

    Conference30th WIC Symposium on Information Theory in the Benelux 2009
    CountryNetherlands
    CityEindhoven
    Period28/05/0929/05/09

    Keywords

    • IR-70158
    • METIS-265748
    • Density estimation
    • EWI-15685
    • eigenvalue estimation
    • SCS-Safety
    • variance correction

    Cite this

    Hendrikse, A. J., Veldhuis, R. N. J., & Spreeuwers, L. J. (2009). Improved variance estimation along sample eigenvectors. In Proceedings of the 30th Symposium on Information Theory in the Benelux (pp. 25-32). Eindhoven: Werkgemeenschap voor Informatie- en Communicatietheorie (WIC).