Training large-scale linear classifiers

--------------------------------------------------------------------
                ***Joint Seminar***
--------------------------------------------------------------------
The Hong Kong University of Science and Technology

Department of Computer Science and Engineering
Human Language Technology Center
--------------------------------------------------------------------

Speaker:	Prof. Chih-Jen LIN
		Department of Computer Science
		National Taiwan University, Taiwan

Title:		"Training large-scale linear classifiers"

Date: 		Thursday, 5 February, 2009

Time: 		10:30am - 11:30am

Venue: 		Lecture Theatre B
		Lam Woo Lecture Theater
		Chia-Wei Woo Academic Concourse, HKUST


Abstract:

In document classification and NLP applications, data often appear
in a rich dimensional feature space. With so many features we do
not need techniques like kernel methods to nonlinearly map data
to a high dimensional space. Instead, linear classifiers are very
suitable for these applications. As data stay in the original input
space, we can train much larger data sets. In this talk, we describe
recent research advances for efficiently training linear SVM, logistic
regression, and maximum entropy. We then discuss future challenges in
handling extremely large data.

******************
Biography:

Chih-Jen Lin is currently a professor in the Department of Computer
Science at National Taiwan University. His research interests include
machine learning, data mining, and related applications. He is best
known for his work on Support Vector Machines (SVMs), a supervised
learning technique. His LIBSVM package is probably the most widely
used implementation of SVMs.