Feature decomposition and structural learning for multi-diverse and multi-view data clustering

In recent years, real-world data often encompass multiple views or features, making multi-view clustering into the spotlight of research. Despite significant progress in existing multi-view clustering methods, they still encounter several challenges: (1) Current methods often grapple with high compu...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:The Visual computer Jg. 41; H. 6; S. 4301 - 4320
Hauptverfasser: Zhang, Yong, Liu, Da, Jiang, Li, Wang, Huibing, Liu, Wenzhe
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2025
Springer Nature B.V
Schlagworte:
Artificial Intelligence
Clustering
Computer Graphics
Computer Science
Constraints
Datasets
Decomposition
Feature decomposition
Image Processing and Computer Vision
Learning
Methods
Neural networks
Regularization
Representations
Binary information
Multi-view clustering
Latent representation learning
Flexible feature decomposition
ISSN:0178-2789, 1432-2315
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In recent years, real-world data often encompass multiple views or features, making multi-view clustering into the spotlight of research. Despite significant progress in existing multi-view clustering methods, they still encounter several challenges: (1) Current methods often grapple with high computational complexity, limiting their applicability to large-scale datasets. (2) Most methods lack guided binary information, hampering their ability to learn correlations among multiple views. (3) Directly handling nonlinear structures proves challenging for the majority of methods. To tackle these challenges, this paper proposes a feature decomposition and structural learning for multi-diverse and multi-view data clustering ( FDSL _ M 2 C ). FDSL _ M 2 C utilizes a flexible feature decomposition approach to extract latent and consensus representations from distinct views. Additionally, it incorporates pseudo-label constraints to refine the consensus representations for intra-view similarity learning. Meanwhile, Laplacian regularization constraints are imposed on the latent representations to capture nonlinear structures. Furthermore, FDSL _ M 2 C integrates latent representation learning and clustering into one-step for improved efficiency. Additionally, to enhance the model’s generalization capability, the proposed method introduces Tikhonov regularization as a penalty term for all learned matrices within the model. Extensive experiments on eight widely used datasets demonstrate that our approach outperforms similar methods in this field. Our code is released on https://github.com/lab-807/FDSL_M2C .
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0178-2789
1432-2315
DOI:10.1007/s00371-024-03661-3