Semi-supervised classification learning by discrimination-aware
manifold regularization
Yunyun Wang
a,b
, Songcan Chen
b,
n
, Hui Xue
c
, Zhenyong Fu
a
a
Department of Computer Science and Engineering, Nanjing University of Posts & Telecommunications, Nanjing 210046, PR China
b
Department of Computer Science and Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, PR China
c
School of Computer Science and Engineering, Southeast University, Nanjing 210096, PR China
article info
Article history:
Received 13 December 2013
Received in revised form
20 April 2014
Accepted 23 June 2014
Communicated by Feiping Nie
Available online 30 June 2014
Keywords:
Semi-supervised classification
Manifold regularization
Discrimination
Unsupervised clustering
abstract
Manifold regularization (MR) provides a powerful framework for semi-supervised classification (SSC)
using both the labeled and unlabeled data. It first constructs a single Laplacian graph over the whole
dataset for representing the manifold structure, and then enforces the smoothness constraint over such
graph by a Laplacian regularizer in learning. However, the smoothness over such a single Laplacian graph
may take the risk of ignoring the discrimination among boundary instances, which are very likely from
different classes though highly close to each other on the manifold. To compensate for such deficiency,
researches have already been devoted by taking into account the discrimination together with the
smoothness in learning. However, those works are only confined to the discrimination of the labeled
instances, thus rather limited in boosting the semi-supervised learning. To mitigate such an unfavorable
situation, we attempt to discover the possible discrimination in the available instances first by
performing some unsupervised clustering over the whole dataset, and then incorporate it into MR to
develop a novel discrimination-aware manifold regularization (DAMR) framework. In DAMR, instances
with high similarity on the manifold will be restricted to share the same class label if belonging to the
same cluster, or to have different class labels, otherwise. Our empirical results show the competitiveness
of DAMR compared to MR and its variants likewise incorporating the discrimination in learning.
& 2014 Elsevier B.V. All rights reserved.
1. Introduction
In many real applications, the unlabeled data can be easily and
cheaply collected, while the acquisition of labeled data is usually
quite expensive and time-consuming, especially involving manual
effort. For instance, in web page recommendation, huge amounts
of web pages are available, but few users are willing to spend time
marking which web pages they are interested in. In spam email
detection, a large number of emails can be automatically collected,
yet few of them have been labeled spam or not by users. Conse-
quently, semi-supervised learning, which exploits a large amount
of unlabeled data jointly with the limited labeled data for learning,
has attracted intensive attention during the past decades. In this
paper, we focus on semi-supervised classification, and so far, lots
of semi-supervised classification methods have been developed
[1–4].
Generally, semi-supervised classification methods attempt
to exploit the intrinsic data distribution information disclosed
by the unlabeled data in learning, and the information is usually
considered to be helpful for learning. To exploit the unlabeled
data, some assumption should be adopted for learning. Two
common assumptions in semi-supervised classi fication are the
cluster assumption and the manifold assumption [3–5]. The
former assumes that similar instances are likely to share the same
class label, thus guides the classification boundary passing through
the low density region between clusters. The latter assumes that
the data are resided on some low dimensional manifold repre-
sented by a Laplacian graph, and similar instances should share
similar classification outputs according to the graph. Almost all
off-the-shelf semi-supervised classification methods adopt one or
both of those assumptions explicitly or implicitly [1,4]. For instance,
the large margin semi-supervised classification methods, such as
transductive Support Vector Machine (TSVM) [6],semi-supervised
SVM (S3VM) [7] and their variants [8,9], adopt the cluster assump-
tion. The graph-based semi-supervised classification methods, such
as label propagation [10,11],graphcuts[12] and manifold regular-
ization (MR) [13], adopt the manifold assumption. Furthermore,
there are also methods combining both assumptions for better
performances, such as RegBoost [14] and SemiBoost [15],etc.
In this paper, we concentrate on the MR framework [13], which
provides an effective way for semi-supervised classification [16],
and has been applied in diverse applications such as image
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/neucom
Neurocomputing
http://dx.doi.org/10.1016/j.neucom.2014.06.059
0925-2312/& 2014 Elsevier B.V. All rights reserved.
n
Corresponding author. Tel.: þ86 25 84892956; fax: þ86 25 84892811.
E-mail address: s.chen@nuaa.edu.cn (S. Chen).
Neurocomputing 147 (2015) 299–306