# Applied Math Seminar: The kernel trick for support vector machines

Speaker: Nathan Albin (K-State)

Abstract: In this talk, I will continue on from Pietro's presentations on linear classifiers to discuss what can be done if we need to separate clouds of points that cannot be easily divided by a straight plane. (Think, for example, of a spherical cloud of "red" points that should be separated from a surrounding spherical shell of "blue" points.) I'll briefly review the primal and dual formulations for the hard-margin and soft-margin linear classifiers, show what happens under nonlinear changes of variables, and use Mercer's theorem to explain the "kernel trick" and why it works. (I may even invoke Gershgorin's circle theorem at some point.)

Friday, October 25, 2019 at 3:30pm to 4:20pm

- Department
- Mathematics, Department of
- Subscribe
- Google Calendar iCal Outlook