A Survey on Heterogeneous Transfer Learning

PhD Qualifying Examination


Title: "A Survey on Heterogeneous Transfer Learning"

by

Miss Ying WEI


Abstract:

Transfer learning algorithms have been proposed to improve the learning 
performance of the target domain where usually labelled data are scarce, under 
the help of the source domain where we have a large amount of labelled data. 
The source domain and target domain usually are not the same, otherwise the 
problem degenerates to traditional machine learning. As transfer learning 
defines, the discrepancy between the target and source domain could be any of 
the data distribution, feature space, label space, and predictive function 
mismatches. In this survey, we focus on the case where the source and target 
domain lie in different feature spaces or label spaces. Yang et al. [60] 
initiated to name the setting as "heterogeneous transfer learning". In the big 
data era, heterogeneity is prevalent given the boom of varieties of data, such 
as images, audio, text and so on. Heterogeneous transfer learning enables 
knowledge transfer among these data sources which probably lie in 
incommensurable feature spaces or disparate label spaces. To the best of my 
knowledge, this survey is the first to systematically review related work on 
heterogeneous transfer learning. We discuss the relationship between 
heterogeneous transfer learning and previous transfer learning. Besides, we 
investigate heterogeneous transfer learning's performances on three tasks, 
i.e., transfer for classification, transfer for clustering and transfer for 
understanding. We also present and categorize a bunch of techniques that are 
frequently employed by heterogeneous transfer learning. Finally, we examine a 
list of applications that heterogeneous transfer learning already or 
potentially pays off.


Date:  			Tuesday, 8 November 2016

Time:                  	2:30pm - 4:30pm

Venue:                  Room 3501
                         Lifts 25/26

Committee Members:	Prof. Qiang Yang (Supervisor)
 			Prof. Prof. Nevin Zhang (Chairperson)
 			Prof. Lei Chen
 			Dr. Yangqiu Song


**** ALL are Welcome ****