摘要
针对大部分基于表示学习的多视角聚类算法分裂了表示学习和聚类任务,导致学习的共性表示对聚类任务缺乏针对性的问题,因此提出了一种新的基于伪标签增强的多视角模糊聚类算法。首先,利用多视角非负矩阵分解提取各个视角间的共性表示,并将其融入聚类过程,使学习的共性表示对聚类更具有针对性。其次,引入伪标签学习以提高表示学习的质量。最后,将伪标签增强的表示学习和模糊聚类划分矩阵学习整合在同一个联合学习框架下,实现相互促进。实验表明,所提出的方法的聚类性能优于现有的多种多视角聚类算法。
To address the problem that most multi-view clustering algorithms based on representation learning split the representation learning and clustering tasks,which leads to the lack of relevance of the learned common representations for the clustering,a new multi-view fuzzy clustering algorithm based on pseudo-label enhancement is proposed.Firstly,a multiview nonnegative matrix factorization is used to extract the common representations between each view and incorporate them into the clustering process so that the learned common representations are more relevant for clustering.Secondly,pseudo-label learning is introduced to enhance the quality of representation learning,and finally,pseudo-label enhanced representation learning and fuzzy clustering division matrix learning are integrated under the same joint learning framework to facilitate each other.Experiments show that the clustering performance of the proposed method is superior to some existing multiple multiview clustering algorithms.
作者
杨鸿潭
YANG Hongtan(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi Jiangsu 214122,China)
出处
《信息与电脑》
2023年第8期120-123,127,共5页
Information & Computer
关键词
多视角
伪标签学习
表示学习
模糊聚类
multi-view
pseudo-label learning
representations learning
fuzzy clustering
