没有合适的资源?快使用搜索试试~ 我知道了~
Adaptive Feature Projection With Distribution Alignment for Deep
需积分: 0 0 下载量 23 浏览量
2023-11-30
21:34:04
上传
评论 1
收藏 33.34MB PDF 举报
温馨提示
试读
13页
Adaptive Feature Projection With Distribution Alignment for Deep IMVC Incomplete Multi-View Clustering
资源推荐
资源详情
资源评论
1354 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 32, 2023
Adaptive Feature Projection With Distribution
Alignment for Deep Incomplete
Multi-View Clustering
Jie Xu , Chao Li, Liang Peng , Yazhou Ren , Member, IEEE, Xiaoshuang Shi ,
Heng Tao Shen , Fellow, IEEE, and Xiaofeng Zhu , Senior Member, IEEE
Abstract— Incomplete multi-view clustering (IMVC) analysis,
where some views of multi-view data usually have missing data,
has attracted increasing attention. However, existing IMVC meth-
ods still have two issues: 1) they pay much attention to imputing
or recovering the missing data, without considering the fact that
the imputed values might be inaccurate due to the unknown
label information, 2) the common features of multiple views are
always learned from the complete data, while ignoring the feature
distribution discrepancy between the complete and incomplete
data. To address these issues, we propose an imputation-free
deep IMVC method and consider distribution alignment in
feature learning. Concretely, the proposed method learns the
features for each view by autoencoders and utilizes an adaptive
feature projection to avoid the imputation for missing data.
All available data are projected into a common feature space,
where the common cluster information is explored by maximizing
mutual information and the distribution alignment is achieved
by minimizing mean discrepancy. Additionally, we design a new
mean discrepancy loss for incomplete multi-view learning and
make it applicable in mini-batch optimization. Extensive experi-
ments demonstrate that our method achieves the comparable or
superior performance compared with state-of-the-art methods.
Index Terms— Incomplete multi-view clustering, adaptive fea-
ture projection, distribution alignment, deep feature learning.
Manuscript received 25 April 2022; revised 7 November 2022 and 4 January
2023; accepted 27 January 2023. Date of publication 13 February 2023;
date of current version 22 February 2023. This work was supported in part
by the National Key Research and Development Program of China under
Grant 2022YFA1004100 and in part by the Medico-Engineering Cooperation
Funds from the University of Electronic Science and Technology of China
under Grant ZYGX2022YGRH009 and Grant ZYGX2022YGRH014. The
associate editor coordinating the review of this manuscript and approv-
ing it for publication was Dr. Claudio R. Jung. (Corresponding authors:
Xiaofeng Zhu; Heng Tao Shen.)
Jie Xu, Chao Li, Liang Peng, Xiaoshuang Shi, and Heng Tao Shen are with
the School of Computer Science and Engineering, University of Electronic
Science and Technology of China, Chengdu 611731, China (e-mail:
jiexuwork@outlook.com; lichao.cfm@gmail.com; larrypengliang@
gmail.com; xsshi2013@gmail.com; shenhengtao@hotmail.com).
Yazhou Ren is with the School of Computer Science and Engineering,
University of Electronic Science and Technology of China, Chengdu 611731,
China, and also with the Shenzhen Institute for Advanced Study, University
of Electronic Science and Technology of China, Shenzhen 518000, China
(e-mail: yazhou.ren@uestc.edu.cn).
Xiaofeng Zhu is with the School of Computer Science and Engineering,
University of Electronic Science and Technology of China, Chengdu 611731,
China, also with the Shenzhen Institute for Advanced Study, University of
Electronic Science and Technology of China, Shenzhen 518000, China, and
also with the Guangxi Academy of Sciences, Nanning 530007, China (e-mail:
seanzhuxf@gmail.com).
This article has supplementary downloadable material available at
https://doi.org/10.1109/TIP.2023.3243521, provided by the authors.
Digital Object Identifier 10.1109/TIP.2023.3243521
I. INTRODUCTION
M
ULTI-VIEW data widely exist in real-world applica-
tions [1], [2], [3], [4], [5], [6], [7], where each sample
consists of multiple views/modalities extracted by different
sensors or different preprocessing methods [8], [9], [10],
[11], [12], [13], [14]. As an important unsupervised learning
approach, multi-view clustering analysis aims to explore the
common cluster patterns among multi-view data and has been
widely investigated in machine learning and image processing
communities [15], [16], [17], [18], [19], [20], [21].
In real-world scenarios, however, the data collection is
easily incomplete for multiple views, where the data are
partially missing in some views. For example, video data
can simultaneously contain multiple views such as images,
captions, and sounds, but some of videos lack captions or
sounds. Unfortunately, existing multi-view clustering methods
are inapplicable to the incomplete multi-view data, thereby
raising the currently attention-getting incomplete multi-view
clustering (IMVC) problem [22]. The goal of IMVC is to
discover the common cluster patterns hidden in incomplete
multi-view data, and previous IMVC methods can be roughly
divided into two groups, ie, traditional and deep methods.
Traditional IMVC methods can be further classified into
four categories: kernel based IMVC methods [23], [24], non-
negative matrix factorization based IMVC methods [22], [25],
[26], tensor based IMVC methods [27], [28], [29], and graph
based IMVC methods [30], [31], [32]. They usually tend to
impute/recover/infer values for the missing data of incomplete
multi-view data sets, and then explore cluster information.
However, traditional IMVC methods have limited capability
to learn feature representations, and meanwhile, some of them
consume high computation costs (e.g., in inverse operations of
matrices) [33].
In recent years, deep IMVC methods (e.g., [34], [35]) have
achieved impressive progresses by combining clustering with
the representation learning of deep models, e.g., deep autoen-
coders. Generally, deep methods leverage the generalization
capability of deep models to achieve the imputation for miss-
ing data. For instance, Lin et al. [36] reconstruct the missing
data with additional prediction networks in a contrastive learn-
ing framework. Some works combine discriminator networks
with autoencoders and employ adversarial training to generate
possible values for the missing data [33], [37].
1941-0042 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Tsinghua University. Downloaded on November 25,2023 at 01:00:29 UTC from IEEE Xplore. Restrictions apply.
XU et al.: ADAPTIVE FEATURE PROJECTION WITH DISTRIBUTION ALIGNMENT FOR DEEP IMVC 1355
Fig. 1. The framework of APADC. (a) APADC learns the view-specific features [Z
v
C
; Z
v
I
] by optimizing the reconstruction loss L
v
R EC
of individual
autoencoder and avoids the imputation for missing data via the adaptive feature projection F . (b) Based on an indicator matrix A and a weight matrix W,
F obtains the common features Z for clustering by mapping {[Z
v
C
; Z
v
I
]}
V
v=1
into a common feature space, where (c) the mutual information loss L
M M I
is optimized to explore the common cluster information contained in {Z
v
C
}
V
v=1
, meanwhile, the mean discrepancy loss L
M M D
is designed to align the
distributions between Z and {[Z
v
C
; Z
v
I
]}
V
v=1
.
However, previous IMVC methods still have two issues as
follows. (1) Most of the existing IMVC methods, including
the traditional and deep methods, incorporate the imputation
process for missing data to complete the incomplete multi-
view data [35]. The imputation of missing data is time-
consuming, moreover, its precision depends on the estimation
of the distribution of available data. As incomplete multi-view
data usually have biased data distribution, the imputation of
missing data might produce noise even incorrect information,
thereby limiting the effectiveness of the imputation models.
(2) Many IMVC methods handle the incomplete multi-view
data in a two-stage process, i.e., first exploring the consistency
among multiple views from the complete part of data, and then
extending the learned consistency to the incomplete part of
data. For example, the methods [33], [36] learn cluster infor-
mation of multiple views on the complete multi-view data, and
then produce imputation values to fill the incomplete multi-
view data. However, feature learning performed on partial data
might cause the distribution discrepancy between the features
of complete data and that of incomplete data, resulting in the
degradation of model generalization capability.
To address the aforementioned issues, we propose APADC:
Adaptive feature Projection with distribution Alignment for
Deep incomplete multi-view Clustering. The proposed method
is an imputation-free framework shown in Fig. 1(a), which
consists of multiple deep autoencoders (to learn the view-
specific features for individual views) and an adaptive fea-
ture projection (to obtain the common features shared by
all views). Specifically, to avoid the imputation of missing
data, we present the adaptive feature projection as shown in
Fig. 1(b), which produces the common features by projecting
the features of all views’ available data into a common feature
space. In particular, an indicator matrix and a weight matrix
are embedded in the adaptive feature projection, to make it
adaptive to varying amounts of missing data and provide dif-
ferent weights to the views according to their feature qualities.
For the feature learning in Fig. 1(c), we maximize the mutual
information between the common features and the view-
specific features of complete multi-view data to explore their
common cluster information. Simultaneously, we minimize the
mean discrepancy to align the feature distributions between
the complete and incomplete data. In this way, the common
features obey the same distribution, which is conducive to
discovering the common cluster patterns across multiple views.
Compared to previous IMVC methods, the main contribu-
tions of our method are summarized as follows.
• We propose an imputation-free deep IMVC method via an
adaptive feature projection, which avoids the imputation
process for missing data so as to prevent noise generation
when handling incomplete multi-view data.
• Our method explores the common cluster information
hidden in multi-view data as well as aligns the distribu-
tions between the complete and incomplete data, which
is usually neglected in existing IMVC methods.
• We design a novel mean discrepancy loss for incomplete
multi-view learning and make it applicable in mini-batch
optimization. Extensive experiments on five popular data
sets indicate that our method obtains comparable or supe-
rior performance over recent state-of-the-art methods.
II. RELATED WORK
A. Deep Multi-View Clustering
Inspired by the success of deep learning, people proposed a
number of deep models recently for multi-view clustering and
obtained impressive achievement [38], [39]. One of the most
popular network architectures in deep multi-view clustering is
autoencoder, which learns features or representations by opti-
mizing the reconstruction loss between the input and the out-
put [40], [41]. Deep autoencoders usually are combined with
existing clustering objectives to perform multi-view clustering.
For instance, Abavisani et al. [38] proposed a multi-view
Authorized licensed use limited to: Tsinghua University. Downloaded on November 25,2023 at 01:00:29 UTC from IEEE Xplore. Restrictions apply.
1356 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 32, 2023
subspace clustering method based on the latent representations
learned by autoencoders. Li et al. [42] incorporated the
adversarial training [43] with autoencoders in a unified multi-
view clustering framework. Fan et al. [44] introduced graph
autoencoders with graph constraints for multi-view clustering.
Yin et al. [45] proposed variational autoencoder based multi-
view clustering, where the shared features were learned with a
mixture of Gaussian distributions. Xu et al. [46] pointed out
to learn disentangled multi-view representations for clustering
with common and peculiar variables in variational autoen-
coders. Different from using the reconstruction objectives of
autoencoders, some works (e.g., [47], [48], [49]) proposed to
penalize the representation space with regularization constrains
for deep multi-view clustering. For example, Zhou et al. [47]
utilized encoder networks to extract informative features and
leveraged Gaussian kernel matrices to avoid the feature degen-
eration.
However, the real-word multi-view data always contain
missing data in some views, resulting in the inapplicability
of existing multi-view clustering methods. Therefore, deep
incomplete multi-view clustering is an important topic which
has attracted researchers’ attention in recent years.
B. Incomplete Multi-View Clustering
Incomplete multi-view clustering (IMVC) is also called par-
tial multi-view clustering in the literature. Traditional IMVC
methods utilize classic machine learning techniques such as
non-negative matrix factorization, kernel trick, graph learning,
and tensor techniques. Li et al. [22] proposed the non-negative
matrix factorization based method to handle incomplete multi-
view data. Hu et al. [26] incorporated weighted and regular-
ized matrix factorization in an online IMVC framework. The
recent matrix factorization based method [50] employed the
cosine similarity to preserve the manifold structures. Matrix
factorization based IMVC usually recovers the non-negative
matrix for the missing data with the available data. Similarly,
kernel based IMVC usually imputes the kernel matrix of
incomplete multi-view data by utilizing that of complete
multi-view data. For example, Guo et al. [23] presented a
kernel similarity based method with an anchor strategy for
partial multi-view clustering. Liu et al. [24] proposed multi-
ple kernel IMVC method, which completed each incomplete
base matrix of incomplete views with the learned consensus
matrix. The graph based IMVC is able to leverage graph
structure information to improve the recognition ability for
cluster patterns. For instance, the literature [31] established
graph regularization to achieve the consistency between the
available data and the imputed values for missing data. The
recent graph based method [51] considered the instance-to-
anchor and instance-to-instance similarities for spectral clus-
tering. Fang et al. [52] leveraged the biological evolution
theory to handle the unbalanced incompleteness in IMVC.
Recent works of tensor based IMVC (e.g., [27], [28], [29])
usually introduce low-rank tensor constraints to characterize
the high-order correlation and the inner structure among
multiple views. In recent years, deep learning based IMVC
has been attracting increasing attention. One of the natural
motivations is that the generative adversarial net (GAN [43])
can be applied to generate data for incomplete multi-view
data [33], [37]. Besides, Wen et al. [35] proposed a cognitive
deep incomplete multi-view clustering network, where the
nearest neighbor graph was constructed and the missing data
was filled by average values. Wei et al. [34] utilized shared
subspace representations to reconstruct the missing data via
a decoder of individual view. Recent work [36] stacked dual
prediction networks on autoencoders to perform data recovery
for incomplete data. Xu et al. [53] proposed to mine the non-
linear cluster complementarity among the incomplete multi-
view data. Tang et al. [54] proposed to dynamically impute
missing views with the learned semantic neighbors.
Most of the traditional and deep IMVC methods han-
dle incomplete multi-view data with imputation/recovery/
inference strategies. However, the inaccurate imputation values
for missing data will negatively affect the performance. This
issue is likely to occur when the number of missing data
is large. Additionally, previous IMVC methods usually learn
the common representations from the complete multi-view
data and generalize them to incomplete multi-view data. This
process might cause the distribution discrepancy between the
complete data and incomplete data. In this paper, we propose
an imputation-free deep IMVC method by considering dis-
tribution alignment in feature learning to address the above
issues.
III. METHOD
Notations: In this paper, {X
v
∈ R
N ×D
v
}
V
v=1
represents a
multi-view data set with V views, where D
v
is the dimen-
sionality of samples in the v-th view and N is the number
of samples. Moreover, we employ an indicator matrix A ∈
{0, 1}
N ×V
, where a
iv
∈ A, a
iv
= 0 denotes that the data of the
i-th sample in the v-th view is missing, and a
iv
= 1 represents
that data is not missing. Denoting the complete data of all
views as {X
v
C
}
V
v=1
and the incomplete data of individual view
as X
v
I
, respectively, for each x
v
i
, if there exists
P
V
v=1
a
iv
= V ,
then x
v
i
∈ X
v
C
; otherwise, x
v
i
∈ X
v
I
. Therefore, [X
v
C
; X
v
I
] ∈
R
N
v
×D
v
and the missing data result in N
v
≤ N . Table I lists
the defined notations and descriptions.
A. Motivation and Framework
Deep autoencoder has been widely applied in IMVC meth-
ods due to its ability of learning clustering-friendly fea-
tures [33], [35], [36]. Concretely, these methods optimize the
reconstruction loss L
R EC
of all views by
L
R EC
=
V
X
v=1
L
v
R EC
=
V
X
v=1
X
v
− D
θ
v
(E
ψ
v
(X
v
))
2
F
, (1)
where E
ψ
v
and D
θ
v
denote the encoder and decoder net-
works of the v-th view, respectively. The encoder network
converts the raw data [X
v
C
; X
v
I
] into the view-specific features
[Z
v
C
; Z
v
I
] ∈ R
N
v
×L
to learn underlying characteristics, i.e.,
[Z
v
C
; Z
v
I
] = E
ψ
v
([X
v
C
; X
v
I
]). (2)
Authorized licensed use limited to: Tsinghua University. Downloaded on November 25,2023 at 01:00:29 UTC from IEEE Xplore. Restrictions apply.
剩余12页未读,继续阅读
资源评论
麻辣小凉皮
- 粉丝: 100
- 资源: 4
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- Docker技术:Docker安装与配置教程+运维技术+超融合+虚拟技术+云计算
- ZeRO Memory Optimizations Toward Training LLM.pdf
- 高效SQL语句编写(how-to-write-efficient-sql)
- ZeroTermux-release_sign.apk
- 非弹性斜碰撞物理课件模拟-HTML网页制作
- 软件测试学习日志-测试基础-day02
- 制作一个简单的进销存(库存管理)页面.rar
- Mixture-of-Experts with Expert Choice Routing.pdf
- Mixtral AI.pdf
- Learning Factored Representations in a Deep MOEs.pdf
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功