Algorithms for non-linear M-estimation

Kaj Madsen, O Edlund, H Ekblom

    Research output: Contribution to journalJournal articleResearchpeer-review


    In non-linear regression, the least squares method is most often used. Since this estimator is highly sensitive to outliers in the data, alternatives have became increasingly popular during the last decades. We present algorithms for non-linear M-estimation. A trust region approach is used, where a sequence of estimation problems for linearized models is solved. In the testing we apply four estimators to ten non-linear data fitting problems. The test problems are also solved by the Generalized Levenberg-Marquardt method and standard optimization BFGS method. It turns out that the new method is in general more reliable and efficient
    Original languageEnglish
    JournalComputational Statistics
    Issue number3
    Pages (from-to)373-383
    Publication statusPublished - 1997

    Fingerprint Dive into the research topics of 'Algorithms for non-linear M-estimation'. Together they form a unique fingerprint.

    Cite this