Systems Seminar
Learning on Manifolds
Dr. Misha Belkin
Abstract
The concept of a Riemannian manifold provides a very general notion of
nonlinear dependency in the data. I will discuss how different
problems of machine learning may be approached in the manifold setting
by reconstructing the Laplace operator on the manifold. In particular,
I will talk about the problem of semi-supervised learning which seems
particularly suited for the manifold methods and will present some
encouraging experimental results. Certain theoretical convergence
guarantees will also be discussed.
This is joint work with Partha Niyogi.
Time and Place: Wed., Mar. 31, at 3:30 pm in 4610 EH
SYSTEMS SEMINAR WEB PAGE:
http://www.cae.wisc.edu/~gubner/seminar/