Sparse on-line Gaussian processes

Lehel Csató, Manfred Opper

    Research output: Working paperTechnical report

    Abstract

    We develop an approach for sparse representations of Gaussian Process (GP) models (which are Bayesian types of kernel machines) in order to overcome their limitations for large data sets. The method is based on a combination of a Bayesian online algorithm together with a sequential construction of a relevant subsample of the data which fully specifies the prediction of the GP model. By using an appealing parametrisation and projection techniques that use the RKHS norm, recursions for the effective parameters and a sparse Gaussian approximation of the posterior process are obtained. This allows both for a propagation of predictions as well as of Bayesian error measures. The significance and robustness of our approach is demonstrated on a variety of experiments.
    Original languageEnglish
    Place of PublicationBirmingham
    PublisherAston University
    Number of pages18
    VolumeNCRG/2001/014
    Publication statusPublished - 9 Sep 2002

    Fingerprint

    Experiments

    Keywords

    • sparse representations
    • Gaussian Process
    • large data sets
    • online algorithm

    Cite this

    Csató, L., & Opper, M. (2002). Sparse on-line Gaussian processes. Birmingham: Aston University.
    Csató, Lehel ; Opper, Manfred. / Sparse on-line Gaussian processes. Birmingham : Aston University, 2002.
    @techreport{9efa4acdaf904bb0b926e0e1923125d1,
    title = "Sparse on-line Gaussian processes",
    abstract = "We develop an approach for sparse representations of Gaussian Process (GP) models (which are Bayesian types of kernel machines) in order to overcome their limitations for large data sets. The method is based on a combination of a Bayesian online algorithm together with a sequential construction of a relevant subsample of the data which fully specifies the prediction of the GP model. By using an appealing parametrisation and projection techniques that use the RKHS norm, recursions for the effective parameters and a sparse Gaussian approximation of the posterior process are obtained. This allows both for a propagation of predictions as well as of Bayesian error measures. The significance and robustness of our approach is demonstrated on a variety of experiments.",
    keywords = "sparse representations, Gaussian Process, large data sets, online algorithm",
    author = "Lehel Csat{\'o} and Manfred Opper",
    year = "2002",
    month = "9",
    day = "9",
    language = "English",
    volume = "NCRG/2001/014",
    publisher = "Aston University",
    type = "WorkingPaper",
    institution = "Aston University",

    }

    Csató, L & Opper, M 2002 'Sparse on-line Gaussian processes' Aston University, Birmingham.

    Sparse on-line Gaussian processes. / Csató, Lehel; Opper, Manfred.

    Birmingham : Aston University, 2002.

    Research output: Working paperTechnical report

    TY - UNPB

    T1 - Sparse on-line Gaussian processes

    AU - Csató, Lehel

    AU - Opper, Manfred

    PY - 2002/9/9

    Y1 - 2002/9/9

    N2 - We develop an approach for sparse representations of Gaussian Process (GP) models (which are Bayesian types of kernel machines) in order to overcome their limitations for large data sets. The method is based on a combination of a Bayesian online algorithm together with a sequential construction of a relevant subsample of the data which fully specifies the prediction of the GP model. By using an appealing parametrisation and projection techniques that use the RKHS norm, recursions for the effective parameters and a sparse Gaussian approximation of the posterior process are obtained. This allows both for a propagation of predictions as well as of Bayesian error measures. The significance and robustness of our approach is demonstrated on a variety of experiments.

    AB - We develop an approach for sparse representations of Gaussian Process (GP) models (which are Bayesian types of kernel machines) in order to overcome their limitations for large data sets. The method is based on a combination of a Bayesian online algorithm together with a sequential construction of a relevant subsample of the data which fully specifies the prediction of the GP model. By using an appealing parametrisation and projection techniques that use the RKHS norm, recursions for the effective parameters and a sparse Gaussian approximation of the posterior process are obtained. This allows both for a propagation of predictions as well as of Bayesian error measures. The significance and robustness of our approach is demonstrated on a variety of experiments.

    KW - sparse representations

    KW - Gaussian Process

    KW - large data sets

    KW - online algorithm

    M3 - Technical report

    VL - NCRG/2001/014

    BT - Sparse on-line Gaussian processes

    PB - Aston University

    CY - Birmingham

    ER -

    Csató L, Opper M. Sparse on-line Gaussian processes. Birmingham: Aston University. 2002 Sep 9.