Journal of Frontiers of Computer Science and Technology ›› 2022, Vol. 16 ›› Issue (1): 242-252.DOI: 10.3778/j.issn.1673-9418.2009020

• Theory and Algorithm • Previous Articles     Next Articles

Semi-supervised Multi-view Classification via Consistency Constraints

LIU Yu, MENG Min+(), WU Jigang   

  1. School of Computer Science and Technology, Guangdong University of Technology, Guangzhou 510006, China
  • Received:2020-09-08 Revised:2020-11-06 Online:2022-01-01 Published:2020-11-19
  • About author:LIU Yu, born in 1996, M.S. candidate. His research interests include image processing and face recognition.
    MENG Min, born in 1984, Ph.D., associate professor, member of CCF. Her research interests include image processing, machine learning, etc.
    WU Jigang, born in 1963, Ph.D., distinguished professor at School of Computer Science and Technology, Guangdong University of Technology, member of CCF. His research interests include network computing, cloud computing, machine intelligence and reconfigurable architecture.
  • Supported by:
    National Natural Science Foundation of China(61702114);National Natural Science Foundation of China(61672171)

一致性约束的半监督多视图分类

刘宇, 孟敏+(), 武继刚   

  1. 广东工业大学 计算机学院,广州 510006
  • 通讯作者: + E-mail: minmeng@gdut.edu.cn
  • 作者简介:刘宇(1996—),男,湖南衡阳人,硕士研究生,主要研究方向为图像处理、人脸识别。
    孟敏(1984—),女,河南汝南人,博士,副教授,CCF会员,主要研究方向为图像处理、机器学习等。
    武继刚(1963—),男,江苏沛县人,博士,广东工业大学计算机学院特聘教授,CCF会员,主要研究方向为网络计算、云计算、机器智能、可重构体系结构。
  • 基金资助:
    国家自然科学基金(61702114);国家自然科学基金(61672171)

Abstract:

Since the traditional semi-supervised multi-view algorithms seldom take into account the diversity of information contained in different views and neglect the consistency of spatial structure between different views, they hardly achieve promising performance when dealing with multi-view data with noise and outlying entries. Although some researchers have proposed semi-supervised multi-view methods,these methods do not make full use of sample discriminant information and subspace structure information under different metric learning,which leads to the unsatisfactory classification results. To deal with the above problems,this paper proposes a semi-supervised multi-view classification via consistency constraint (SMCC) for multi-view data analysis. Firstly, the consistency constraints between different views are enhanced based on the Hilbert-Schmidt independence criteria (HSIC). Then, the dimensionality reduction is performed by feature projection to preserve the local manifold structure, which is integrated with Frobenius norm constraint to improve the robustness of the algorithm. Furthermore, the corresponding weights are adaptively assigned to different views to reduce the influence of feature information and noise pollution in different views. Finally, the proposed model can be solved efficiently using the linear alternative direction method with adaptive penalty and eigen-decomposition. The experimental results on four benchmark datasets show that the proposed algorithm can discover more effective discriminant information from multi-view data and its accuracy is improved.

Key words: multi-view, adaptive weight, consistency constraints, feature projection, semi-supervised learning

摘要:

由于传统半监督模式下的多视图算法很少考虑到不同视图中数据包含信息的差异性,且忽视了不同视图间存在着空间结构的一致性,算法在含有噪声和异常点的多视图数据中性能较差。尽管有研究者已经提出了半监督多视图方法,但这些方法没有充分利用样本判别信息以及不同度量学习下的子空间结构信息,从而导致分类结果不理想。针对以上问题,提出了一致性约束的半监督多视图分类算法(SMCC)。首先,基于希尔伯特-施密特独立性准则(HSIC)加强对不同视图之间的一致性约束。然后,通过保留原始数据的空间局部流形结构进行特征投影来降低数据空间维度,并结合F范数约束提高算法的鲁棒性。进一步,对不同视图自适应地赋予相应的权重,降低在不同视图中数据含有不同特征信息与噪声污染的影响。最后,基于线性交替方向乘子法与特征分解方法对模型进行求解。在四个基准数据集上的实验结果表明,提出的算法能够捕获多视图数据中更多的有效判别信息,准确性得到了提高。

关键词: 多视图, 自适应权重, 一致性约束, 特征投影, 半监督学习

CLC Number: