Professor Robert Kallman invites you to attend the PhD dissertation defense of Sam McWhorter next Friday, March 21st at 12:15 pm in Auditorium Bldg room 202. Cookies and coffee will be served in GAB 472 following the event.
"Some Practical Results In the Theory Of Support-Vector Machines"
Abstract:
Support-vector machines are a category of geometrically-determined binary classifiers. They are, in principle, capable of expressing any such classification by mapping their inputs into disjoint regions separated by hyperplane in a Hilbert space. Non-linearities in the input are managed by selection of a positive-definite kernel function that implicitly maps the data into geometrically distinct regions. Computations of this type often involve quadratic minimization algorithms. However, Dr. Kallman has created a novel ad-hoc algorithm for support vector machines designed to run more efficiently than general-purpose quadratic minimization algorithms.
In this talk, we will briefly discuss support-vector machines and prove convergence of Dr. Kallman's algorithm. We will also provide a novel elementary proof of the positive-definiteness of a family of functions that include generalizations of the Gaussian radial basis function, a commonly-used kernel.