IEEE Computational Intelligence Magazine - May 2020 - 78

Application
Notes

Lu Liu
School of Journalism and Communication,
Fudan University, Shanghai, CHINA
Yong Luo
School of Computer Science and Engineering,
Nanyang Technological University, SINGAPORE
Han Hu
School of Information and Electronics, Beijing
Institute of Technology, Beijing, CHINA
Yonggang Wen
School of Computer Science and Engineering,
Nanyang Technological University, SINGAPORE
Dacheng Tao
School of Computer Science, The University of
Sydney, Sydney, AUSTRALIA
Xin Yao
Department of Computer Science and Engineering,
Southern University of Science and Technology,
Shenzhen, CHINA

xTML: A Unified Heterogeneous Transfer Metric Learning
Framework for Multimedia Applications
Abstract

O

wing to the continual growth of
multimodal data (or feature spac-
es), we have seen a rising interest
in multimedia applications (e.g., object
classification and searching) over these
heterogeneous data. However, the accu-
racy of classification and searching tasks
is highly dependent on the distance esti-
mation between data samples, and simple
Euclidean (EU) distance has been proven
to be inadequate. Previous research has
focused on learning a robust distance
metric to quantify the relationships
among data samples. In this context,
existing distance metric learning (DML)
algorithms mainly leverage on label
information in the target domain for
model training and may fail when the
label information is scarce. As an
improvement, transfer metric learning
(TML) approaches are proposed to lever-
age information from other related
domains. However, current TML algo-
rithms assume that different domains
explore the same representation; thus,

Digital Object Identifier 10.1109/MCI.2020.2976187
Date of current version: 10 April 2020

78

©ISTOCKPHOTO.COM/NATASAADZIC

they are not applicable in heterogeneous
settings where the data representations of
different domains vary. In this research,
we propose xTML, a novel unified het-
erogeneous transfer metric learning
framework, to improve the distance esti-
mation of the domains of interest (i.e.,
the target domains in classification and
searching tasks) when limited label

Corresponding Author:Yonggang Wen (Email: ygwen@
ntu.edu.sg). X.Yao is also with the School of Computer
Science, University of Birmingham, UK.

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | MAY 2020

information, complementary with ex-
tensive unlabeled data, is provisioned for
model training. We further illustrate
how our proposed framework can be
applied to a selected list of multimedia
applications, including opinion mining,
deception detection and online product
searching. Qualitative and quantitative
comparisons between our proposed
algorithm and alternative solutions
demonstrate the advantages of our pro-
posed framework in multiple perfor-
mance metrics.

1556-603X/20©2020IEEE
http://www.ISTOCKPHOTO.COM/NATASAADZIC
IEEE Computational Intelligence Magazine - May 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - May 2020
