I will introduce a class of flexible (conditional probability) models
and techniques for classification problems. Many existing methods such
as generalized linear models and support vector machines are subsumed
under this class. The flexibility of this class of techniques comes from
the use of kernel functions as in support vector machines. A kernel
captures a mapping of examples into (high dimensional) feature vectors
and allows the classification to be carried out in the feature space
without ever explicitly representing it. This class of methods can be
characterized in several alternative ways and I will discuss some of
these with examples. I will also touch generalization performance of
these methods and, in particular, the construction of appropriate kernel
functions from generative probability models. Some experimental results
will be given in the context of biosequence analysis illustrating the
effectiveness of these techniques.
Date: Thurs., February 26; Time: 4:15-5:30PM; Place: Gates 100
Return to seminar schedule.