A Survey on Transfer Learning

PhD Qualifying Examination


Title: "A Survey on Transfer Learning"

Mr. Jialin Pan


Abstract:

Machine learning technology has achieved significant success in many
areas, especially in the areas of classification, regression and
clustering. Most machine learning algorithms have a common assumption that
the distributions between training data and test data are the same.
However, in many real world applications, the training data may be
out-of-date due to dynamic environmental factors.  In addition, we might
want to use the training data in one task domain to learn prediction
models for use in another domain. In these cases, the distributions
between the training data and the test data may be very different. As a
result, most machine learning based systems need to be retrained for every
new situation they encounter. This requires collecting a large amount of
new training examples, which is very expensive and limiting aspect of
deploying such systems, which is often infeasible.

In recent years, transfer learning techniques are purposed to address this
shortcoming by leveraging knowledge learnt in previous problems to solve
new problems effectively with fewer training examples and less training
time. This survey mainly focuses on reviewing the current work on transfer
learning for classification, regression and clustering problems.
Furthermore, we discuss the relationship between transfer learning and
other related research areas, such as domain adaptation, multi-task
learning and sample selection bias.


Date:     		Friday, 27 June 2008

Time:                   10:00a.m.-12:00noon

Venue:                  Room 3501
			lifts 25-26

Committee Members:      Prof. Qiang Yang (Supervisor)
			Dr. Nevin Zhang (Chairperson)
			Dr. James Kwok
			Dr. Charles Zhang


**** ALL are Welcome ****