Recent Development in Multitask Learning

PhD Qualifying Examination


Title: "Recent Development in Multitask Learning"

Mr. Yu ZHANG


Abstract:

Multitask learning is a machine learning approach that aims at improving the 
generalization performance of a learning task with the help of some other 
related tasks. It is inspired by human learning activities in that people often 
apply the knowledge gained from previous learning tasks to help learn a new 
task. Over the past decade, this learning approach has attracted the attention 
of many researchers, making it a hot research topic in the machine learning 
community. In this survey, we review some major recent development in multitask 
learning. We categorize multitask learning methods into four main classes, 
namely, the common representation approach which seeks to find a common data or 
model representation for multiple tasks, the task regularization approach which 
formulates multitask learning under a regularization framework and represents 
the model assumption as a regularization term in the objective function, the 
task clustering approach which clusters the tasks and assumes that a common or 
similar representation is shared among the tasks within the same cluster, and 
the hierarchical Bayesian approach which is based on hierarchical Bayesian 
modeling in Bayesian statistics. Moreover, we discuss the advantages and 
disadvantages of these four approaches, hoping to provide some guidance on 
selecting the appropriate approach given a target application at hand. Besides 
multitask learning, there also exist other approaches to boosting learning 
performance in situations when labeled training data are scarce, such as 
semi-supervised learning and active learning. Along this line, we introduce 
some work that combines multitask learning with these approaches to achieve 
further improvement in performance. After reviewing the different approaches, 
we briefly present some theoretical results that provide theoretical 
justifications on the benefit of multitask learning. On the application side, 
we review some applications in areas such as computer vision and information 
retrieval. Finally, we discuss some possible directions for research in 
multitask learning.


Date:     		Friday, 16 January 2009

Time:                   2:00p.m.-4:00p.m.

Venue:                  Room 3501
 			lifts 25-26

Committee Members:      Prof. Dit-Yan Yeung (Supervisor)
 			Dr. Nevin Zhang (Chairperson)
 			Dr. Albert Chung
 			Dr. James Kwok


**** ALL are Welcome ****