Deep self-weighted multi-view fuzzy clustering

Multi-view clustering has attracted considerable attention in various fields, such as computer vision and information retrieval. Most existing methods adopt a stepwise strategy to achieve a consistent representation and produce final clusters. However, this strategy neglects label consistency for th...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Knowledge-based systems Jg. 328; S. 114158
Hauptverfasser: Shi, Mei, Zhao, Xiaowei, Yin, Xiaoyan, Xiao, Yun, Guo, Jun
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 25.10.2025
Schlagworte:
Alternating optimization algorithm
Deep learning
Fuzzy clustering
Information entropy
Multi-view learning
Deep learning
Fuzzy clustering
Alternating optimization algorithm
Information entropy
Multi-view learning
ISSN:0950-7051
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Multi-view clustering has attracted considerable attention in various fields, such as computer vision and information retrieval. Most existing methods adopt a stepwise strategy to achieve a consistent representation and produce final clusters. However, this strategy neglects label consistency for the same sample across different views, which results in sub-optimal representations. Furthermore, conventional methods frequently overlook the potential fuzzy membership relationships inherent in multi-view data and predominantly rely on shallow models, which fail to capture the complex properties of data, resulting in unsatisfactory outcomes. To address these challenges, we propose a novel deep self-weighted multi-view fuzzy clustering method that thoroughly explores the intricate view-specific characteristics of data to better represent consensus membership (i.e. consistent representation) between samples and centroids across multiple views. In particular, the method uses deep auto-encoders to non-linearly project samples from each view into corresponding latent spaces in a layer-wise manner. The consensus membership is then shared by samples from the middle and reconstruction layers, thereby reducing discrepancies in soft cluster assignment between the same sample in the latent and original spaces. Without introducing additional parameters, the self-weighted strategy adjusts the contribution of each view to fuzzy clustering. In addition, we adopt entropy regularization to tune the uniformity of the membership and design an alternating optimization algorithm to update all variables. Experimental results demonstrate the superior performance of the proposed method on five datasets (including images, web pages and videos) evaluated using four metrics.
ISSN:0950-7051
DOI:10.1016/j.knosys.2025.114158