Experiments using machine learning to approximate likelihood ratios for mixture models

K. Cranmer, J. Pavez, G. Louppe, W. K. Brooks

    Research output: Contribution to journalArticle

    Abstract

    Likelihood ratio tests are a key tool in many fields of science. In order to evaluate the likelihood ratio the likelihood function is needed. However, it is common in fields such as High Energy Physics to have complex simulations that describe the distribution while not having a description of the likelihood that can be directly evaluated. In this setting it is impossible or computationally expensive to evaluate the likelihood. It is, however, possible to construct an equivalent version of the likelihood ratio that can be evaluated by using discriminative classifiers. We show how this can be used to approximate the likelihood ratio when the underlying distribution is a weighted sum of probability distributions (e.g. signal plus background model). We demonstrate how the results can be considerably improved by decomposing the ratio and use a set of classifiers in a pairwise manner on the components of the mixture model and how this can be used to estimate the unknown coefficients of the model, such as the signal contribution.

    Original languageEnglish (US)
    Article number012034
    JournalJournal of Physics: Conference Series
    Volume762
    Issue number1
    DOIs
    StatePublished - Nov 21 2016

    Fingerprint

    likelihood ratio
    machine learning
    classifiers
    physics
    coefficients
    estimates
    simulation
    energy

    ASJC Scopus subject areas

    • Physics and Astronomy(all)

    Cite this

    Experiments using machine learning to approximate likelihood ratios for mixture models. / Cranmer, K.; Pavez, J.; Louppe, G.; Brooks, W. K.

    In: Journal of Physics: Conference Series, Vol. 762, No. 1, 012034, 21.11.2016.

    Research output: Contribution to journalArticle

    Cranmer, K. ; Pavez, J. ; Louppe, G. ; Brooks, W. K. / Experiments using machine learning to approximate likelihood ratios for mixture models. In: Journal of Physics: Conference Series. 2016 ; Vol. 762, No. 1.
    @article{241de69c440640a59bfaf46ee4a6b272,
    title = "Experiments using machine learning to approximate likelihood ratios for mixture models",
    abstract = "Likelihood ratio tests are a key tool in many fields of science. In order to evaluate the likelihood ratio the likelihood function is needed. However, it is common in fields such as High Energy Physics to have complex simulations that describe the distribution while not having a description of the likelihood that can be directly evaluated. In this setting it is impossible or computationally expensive to evaluate the likelihood. It is, however, possible to construct an equivalent version of the likelihood ratio that can be evaluated by using discriminative classifiers. We show how this can be used to approximate the likelihood ratio when the underlying distribution is a weighted sum of probability distributions (e.g. signal plus background model). We demonstrate how the results can be considerably improved by decomposing the ratio and use a set of classifiers in a pairwise manner on the components of the mixture model and how this can be used to estimate the unknown coefficients of the model, such as the signal contribution.",
    author = "K. Cranmer and J. Pavez and G. Louppe and Brooks, {W. K.}",
    year = "2016",
    month = "11",
    day = "21",
    doi = "10.1088/1742-6596/762/1/012034",
    language = "English (US)",
    volume = "762",
    journal = "Journal of Physics: Conference Series",
    issn = "1742-6588",
    publisher = "IOP Publishing Ltd.",
    number = "1",

    }

    TY - JOUR

    T1 - Experiments using machine learning to approximate likelihood ratios for mixture models

    AU - Cranmer, K.

    AU - Pavez, J.

    AU - Louppe, G.

    AU - Brooks, W. K.

    PY - 2016/11/21

    Y1 - 2016/11/21

    N2 - Likelihood ratio tests are a key tool in many fields of science. In order to evaluate the likelihood ratio the likelihood function is needed. However, it is common in fields such as High Energy Physics to have complex simulations that describe the distribution while not having a description of the likelihood that can be directly evaluated. In this setting it is impossible or computationally expensive to evaluate the likelihood. It is, however, possible to construct an equivalent version of the likelihood ratio that can be evaluated by using discriminative classifiers. We show how this can be used to approximate the likelihood ratio when the underlying distribution is a weighted sum of probability distributions (e.g. signal plus background model). We demonstrate how the results can be considerably improved by decomposing the ratio and use a set of classifiers in a pairwise manner on the components of the mixture model and how this can be used to estimate the unknown coefficients of the model, such as the signal contribution.

    AB - Likelihood ratio tests are a key tool in many fields of science. In order to evaluate the likelihood ratio the likelihood function is needed. However, it is common in fields such as High Energy Physics to have complex simulations that describe the distribution while not having a description of the likelihood that can be directly evaluated. In this setting it is impossible or computationally expensive to evaluate the likelihood. It is, however, possible to construct an equivalent version of the likelihood ratio that can be evaluated by using discriminative classifiers. We show how this can be used to approximate the likelihood ratio when the underlying distribution is a weighted sum of probability distributions (e.g. signal plus background model). We demonstrate how the results can be considerably improved by decomposing the ratio and use a set of classifiers in a pairwise manner on the components of the mixture model and how this can be used to estimate the unknown coefficients of the model, such as the signal contribution.

    UR - http://www.scopus.com/inward/record.url?scp=85002323056&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85002323056&partnerID=8YFLogxK

    U2 - 10.1088/1742-6596/762/1/012034

    DO - 10.1088/1742-6596/762/1/012034

    M3 - Article

    VL - 762

    JO - Journal of Physics: Conference Series

    JF - Journal of Physics: Conference Series

    SN - 1742-6588

    IS - 1

    M1 - 012034

    ER -