Transfer learning with application to multimedia analysis

Speaker:	Dr. Rong YAN
		IBM T. J. Watson Research Center

Title:		"Transfer learning with application to multimedia
		 analysis"

Date:		Wednesday, 14 October, 2009

Time:		11:00am - 12 noon

Venue:		Room 3401 (via lifts. 17/18), HKUST


Abstract:

In this talk, I will present two transfer learning algorithms with
application to multimedia analysis. The first algorithm is called Adaptive
Support Vector Machines, designed for transferring across data domains. It
supports both one-to-one and many-to-one domain adaptation by learning a
SVM-type objective function to update the adapted classifiers. We also
study how to select the most useful classifiers as the basis for
adaptation using meta-level features. Our experiments on multimedia
concept detection across channels show promising performance.

The second algorithm describes a probabilistic transfer learning algorithm
with task-level features to control the task mixture selection in a
hierarchical Bayesian model. It is designed for transferring across
information retrieval tasks. This model allows effective transfer to new
tasks, especially when only limited number or even none of target data are
available. To estimate the model parameters, we develop an empirical Bayes
method based on variational approximation techniques. Our experiments on
TRECVID datasets show that the proposed model achieves significantly
better performance compared with other transfer learning methods.


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

Dr. Rong Yan is a Research Staff Member in Intelligent Information
Management Department at the IBM T. J. Watson Research Center. Dr. Yan
received his M.Sc. (2004) and Ph.D. (2006) degree from Carnegie Mellon
University's School of Computer Science. His research interests include
Multimedia information retrieval, large-scale machine learning, data
mining, video content analysis, and computer vision. Dr. Yan is the
leading designer of the automatic video retrieval system that achieves the
best performance in the world-wide TRECVID evaluation in 2003 / 2005. He
received the Best Paper Runner-Up awards in ACM Multimedia 2004 and ACM
CIVR 2007. Dr. Yan is currently a member of the IBM Research Multimedia
Analysis and Retrieval project, which won the 2008 VideoOlympics People's
Choice Award, and 2004 Wall Street Journal Innovation Award. He was a
member of CMU Informedia project, which won the CMU Newell Award for
Research Excellence in 2003. He has got the IBM Research External
Recognition Award in 2007. Dr. Yan has authored or co-authored 5 book
chapters and more than 60 international conference and journal papers. Dr.
Yan has served or is serving as co-chairs for 7 conferences / workshops
and as a Program Committee member in more than 30 ACM / IEEE conferences.
He is a member of ACM.