[Colloquium] Sindhwani/M.S. Presentation/Nov. 3, 2004

[Margaret Jaffey]

This is an announcement of Vikas Sindhwani's Master's Presentation.

-------------
Date:  Wednesday, November 3, 2004

Time:  2:30 p.m.

Place:  Ryerson 251

M.S. Candidate:  Vikas Sindhwani

M.S. Paper Title:  Kernel Machines for Semisupervised Learning

Abstract:
This thesis proposes a unifying framework for learning from examples in
which the geometric structure of the probability distribution underlying
the data is incorporated within the learning mechanism. Families of
algorithms for semi-supervised learning, clustering and data
representation are derived within this framework. Two specific 
algorithms
are presented: Laplacian Regularized Least Squares and the
Laplacian Support Vector Machines, that are natural semi-supervised
extensions of popular regularization methods in Reproducing Kernel 
Hilbert
Spaces. In addition, several recently proposed transductive methods are
also seen to be special cases of this general approach.  New Representer
theorems are proved that provide a theoretical basis for these 
algorithms.
The problem of out-of-sample extension to novel examples (in Spectral
clustering and graph-based approaches to transductive learning) is
natually resolved in this framework. We present experimental evidence
suggesting that our semi-supervised algorithms are able to use unlabeled
data effectively. Finally, this thesis also includes some ideas for
efficient implementation of these algorithms and extentions of this
framework for invariant learning.

Advisor:  Prof. Partha Niyogi

A draft copy of Vikas Sindhwani's MS Paper will be available soon in Ry 
161A.

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Margaret P. Jaffey				margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 161A)		(773) 702-6011
The University of Chicago		http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-