Grace Wahba's Talk at FoCM (Foundations of Computational Mathematics) conference at the IMA, Minneapolis. August 5-14, 2002

Grace Wahba's Talk at the FoCM (Foundations of Computational Mathematics) conference at the IMA, Minneapolis, August 5-14, 2002

Statistical Model Building and Classification as Optimization Problems in RKHS-Polychotomous Penalized Likelihood and Multicategory Support Vector Machines.

The talk examines the Multicategory Support Vector Machine for classification and discusses its relation with penalized likelihood estimation given polychotomous data. The overheads for this talk are here(ps) (pdf). Related talks were given at Max Planck Institute-Tuebingen and ETH-Zurich augmented by a discussion of the tuning by the GACV here. Two recent reports on the Multicategory Support Vector Machine are: Lee and Lee, TR 1051 (2002) and Lee, Lin and Wahba, TR 1043 (2001). Polychotomous Penalized Likelihood estimates are proposed in: Xiwu Lin, TR 1003 (1998) (also thesis). Overheads for a Shortcourse on Reproducing Kernel Hilbert Spaces are here.

The home page for the Conference is here.

A few (classic) references on smoothing parameter selection for the gaussian white noise regression case: here .

Grace Wahba and students' reports available on the web here.

Grace Wahba and students' reports available on the web here. Some `golden oldies' scanned for the web, including Kimeldorf and Wahba (1971) (the representer theorem) here.

Grace Wahba's list of publications here.

Grace Wahba's home page here.