多视图拉普拉斯正则化驱动的半监督稀疏特征选择提升网络图像标注效率

9 浏览量更新于2024-08-26 1 收藏 1.37MB PDF 举报

本文主要探讨了"基于多视图拉普拉斯正则化的半监督稀疏特征选择"这一领域的研究。半监督学习是一种在机器学习中广泛应用的技术，尤其在大规模未标注数据和少量标注数据共存的情况下，如网络图像标注任务中，它能有效地利用这些信息来提升模型性能。然而，传统的半监督稀疏特征选择方法大多针对单一视角的数据设计，这限制了它们处理多源、多维度数据的能力。多视图学习是近年来研究热点，它强调通过整合来自不同视角或模态的数据来挖掘更全面的信息。在本文中，作者提出了一种创新的方法，即通过融合多个视角的拉普拉斯正则化，旨在解决半监督稀疏特征选择问题。拉普拉斯正则化作为一种图形模型中的技术，能够有效地捕捉数据之间的局部结构信息，这对于特征选择至关重要，因为它可以帮助筛选出与目标变量高度相关的特征。具体来说，该方法首先构建了一个多视图的图模型，将每个数据样本视为图中的节点，不同视角的数据属性作为边的权重，这有助于衡量各个特征之间的相似性或关联度。拉普拉斯矩阵在这种框架下被用于量化局部一致性，使得未标注数据可以间接地通过邻近的有标签数据进行学习。通过添加拉普拉斯惩罚项，算法能够引导特征选择过程趋向于选择那些既能保持数据的全局结构，又能确保在有限的标注信息下具有良好的预测能力的稀疏特征集。文章的贡献在于： 1. 提供了一种新的半监督特征选择框架，能够在处理多视图数据时考虑到各视角间的相互作用。 2. 结合了拉普拉斯正则化，使得模型不仅依赖于少量的有标签样本，还能充分利用大量未标注样本的潜在信息。 3. 研究了如何通过优化算法来实现该模型，确保特征选择的效率和准确性。尽管文章在2014年9月接收并进行了修订，最终在2015年6月接受，这表明其研究成果得到了同行的积极评价。关键词包括"多视图学习"、"拉普拉斯正则化"、"半监督学习"和"稀疏特征选择"，反映出这篇论文的核心内容和研究方向。这篇文章对如何在多视图环境下进行半监督稀疏特征选择进行了深入研究，为实际应用中的大规模数据处理提供了一种有效的方法，特别是在图像识别、文本分类等领域具有重要意义。

Semi-supervised sparse feature selection based on multi-view

Laplacian regularization

☆

Caijuan Shi

a,b,c,

⁎

, Qiuqi Ruan

b,c

, Gaoyun An

b,c

,ChaoGe

College of Information Engineering, North China University of Science and Technology, Tangshan 063009, China

Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China

Beijing Key Laboratory of Advanced Information Science and Network Technology, Beijing 100044, China

abstractarticle info

Article history:

Received 2 September 2014

Received in revised form 7 May 2015

Accepted 12 June 2015

Available online 23 June 2015

Keywords:

Multi-view learning

Laplacian regularization

Semi-supervised learning

Sparse feature selection

Semi-supervised sparse feature selection, which can exploit the large number unlabeled data and small number

labeled data simultaneously, has placed an important role in web image annotation. However, most of the semi-

supervised sparse feature selection methods are developed for single-view data and these methods cannot nat-

urally deal with the multi-view data, though it has shown that leveraging information contained in multiple

views can dramatically improve the feature selection performance. Recently, multi-view learning has obtained

much research attention because it can reveal and leverage the correlated and complementary information be-

tween different views. So in this paper, we apply multi-view learning into semi-supervised sparse feature selec-

tion and propose a semi-supervised sparse feature selection method based on multi-view Laplacian

regularization, namely, multi-view Laplacian sparse feature selection (MLSFS).

MLSFS utilizes mult i-view

Laplacian regularization to boost semi-supervised sparse feature selection performance. A simple iterative meth-

od is proposed to solve the objective function of MLSFS. We apply MLSFS algorithm into image annotation task

and conduct experiments on two web image datasets. The experimental results show that the proposed MLSFS

outperforms the state-of-art single-view sparse feature selection methods.

1. Introduction

Web images, most of which are unlabeled, have shown continuous

explosive growth. As an important means, semi-supervised sparse fea-

ture selection [1–4] has the ability to improve the performance of web

image annotation. It has extensively shown that semi-supervised sparse

feature selection approaches can overcome the drawbacks of supervised

feature selection methods and unsupervised feature selection methods.

On the one hand, semi-supervised sparse feature selection approaches

can save human labor cost for labeling a large amount of training data,

and on the other hand, they can make full use the reliable labeled data

and the accessible unlabeled data simultaneously to improve the sparse

feature selection performance.

Among different semi-supervised learning methods, graph Laplacian

regularization based method is one of the most representative works [5].

The graph Laplacian can determine the geometry of the underlying man-

ifold in Laplacian regularization. Now, the graph Laplacian regularization

based semi-supervised learning has been widely applied into semi-

supervised sparse feature selection [1,2].In[1], Ma et al. have proposed

a structural feature selection with sparsity frame based on graph

Laplacian semi-supervised learning to select features with considering

the correlation between them. In [2], Shi et al. have proposed a semi-

supervised sparse feature selection method based on graph Laplacian

and l

2,1/2

-matrix norm to select more sparse and discriminative features

for image annotation. In this paper, we also exploit graph Laplacian regu-

larization to construct our semi-supervised sparse feature selection frame.

As we know, images are usually represented by dif ferent types of

features, such as color correlogram, wavelet texture, and edge direction

histogram. Each type of features characterizes these image s in one

speciﬁc feature space and has particular physical meaning and statistic

property. Conventionally, the data represented by multiple types of

features are named as multi-view data to distinguish from the single-

view data represented only by one type of features [6]. However, most

of the existing semi-supervised sparse feature selection methods are

developed for the single-view data and these methods concatenate

multiple views featu res into a long vector once they confront with

multi-view data, such as [1,2]. This concatenation strategy cannot efﬁ-

ciently explore the complementary of different view features because

it improperly treats different view features carrying different physical

characteristics. In additio n, this concatenation strategy ignores the

Image and Vision Computing 41 (2015) 1–10

☆

This paper has been recommended for acceptance by Etienne Memin.

⁎ Corresponding author at: College of Information Engineering, North China University

of Science and Technology, Tangshan 063009, China. Tel.: +86 15630555090.

E-mail addresses: shicaijuan2011@gmail.com (C. Shi), qqruan@center.njtu.edu.cn

(Q. Ruan), gyan@bjtu.edu.cn (G. An), chaoge@ncst.edu.cn (C. Ge).

MLSFS: Multi-view Laplacian Sparse Feature Selection.

http://dx.doi.org/10.1016/j.imavis.2015.06.006

Contents lists available at ScienceDirect

Image and Vision Computing

journal homepage: www.elsevier.com/locate/imavis

下载后可阅读完整内容，剩余9页未读，立即下载

weixin_38683930

粉丝: 2
资源: 879

多视图拉普拉斯正则化驱动的半监督稀疏特征选择提升网络图像标注效率

通过混合图拉普拉斯正则化进行渐进图像复原

多视图Hessian正则化下的半监督稀疏特征选择

致力于设计基于风险的安全拉普拉斯正则化最小二乘法

用于多媒体分析的多视图黑森州半监督稀疏特征选择

超拉普拉斯正则化单向低秩张量恢复用于多光谱图像降噪

三维点云去噪技术：基于低维流形图拉普拉斯正则化

风险导向的安全拉普拉斯正则化最小二乘法设计

拉普拉斯正则化概率主元分析在故障检测中的应用

超拉普拉斯正则化单向低秩张量恢复：多光谱图像降噪新方法

LRLS拉普拉斯正则化

最新资源