[xtag-meeting]: Libin on SVMs this week: 06/06/2002

XTAG Meeting: Thursday 10:30am, IRCS Fishbowl 06/06/2002

This week, Libin will give a practice talk. Abstract follows.


Support Vector Machines


Support Vector Machines (SVM) are known to achieve high generalization
performance even with input data of high demensional feature spaces. In
the present paper, we first introduce the definition of SVM and the
approach to solve SVM with Lagrange function (Lagrangian) and KT
conditions. Then kernel functions are intoduced in the non-linear SVMs.
We further analyze the generalized ability of SVMs with VC dimenstion and
other bounds, and exposit the practicality of large-scale SVMs based on
SVM^light. The paper ends with SVM's application in Chunking, an NLP