Learning similarities between classes

Cheng Soon Ong (University of Melbourne)

NICTA SML SEMINAR

DATE: 2012-04-26
TIME: 11:15:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
We propose a new learning problem of learning the class similarity from data. In the process of finding a solution of this problem, we revisit four accepted facts in machine learning: - Finding a good kernel can be done using multiple kernel learning - Convex optimization is nice because local minima are global minima - Solving semi definite programs are efficient - Clustering is a hard problem