Transfer Learning via Feature Clustering

Speaker:	Arthur Wenyuan DAI
		Department of Computer Science and Engineering
		Shanghai Jiaotong University

Title:		"Transfer Learning via Feature Clustering"

Date:		Monday, 6 October 2008

Time:		4:00pm - 5:00pm

Venue:		Lecture Theatre F
		(Leung Yat Sing Lecture Theatre, near lifts 25/26)
		HKUST

Abstract:

Transfer learning focuses on using learned knowledge in one context to
benefit the future learning in other contexts. In this talk, I will show a
transfer learning technique through feature clustering. The feature
clustering can be considered as a new feature representation, under which
the target learning can be enhanced. In our transfer learning technique, the
two learning contexts share a common feature clustering, and thus the
feature clustering serves as a "bridge" to let the learning in one context
influence the learning in another. We show two applications in domain
adaptation and unsupervised transfer clustering. The experimental results
show that our transfer learning technique can outperform several
state-of-the-art learning algorithms in transfer learning.


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

Arthur Wenyuan Dai is a PhD Student at Shanghai Jiaotong University, where
he obtained his BSC and MSC degrees.  He has been active in machine learning
and data mining, focusing in particular on the problem of transfer learning,
where his work has been published at ICML 2007, AAAI 2007, WWW 2008, ICML
2008, NIPS 2008 and SIGIR 2008.  This work has been done with in conjunction
with Professor Qiang Yang, Professor Yong Yu and Dr. Guirong Xue.