Regularization with a pruning prior

Cyril Goutte, Lars Kai Hansen

    Research output: Contribution to journalJournal articleResearchpeer-review

    Abstract

    We investigate the use of a regularization priorthat we show has pruning properties. Analyses areconducted both using a Bayesian framework and withthe generalization method, on a simple toyproblem. Results are thoroughly compared withthose obtained with a traditional weight decay.
    Original languageEnglish
    JournalNeural Networks
    Volume10
    Issue number6
    Pages (from-to)1053-1059
    ISSN0893-6080
    Publication statusPublished - 1997

    Fingerprint

    Dive into the research topics of 'Regularization with a pruning prior'. Together they form a unique fingerprint.

    Cite this