Reducing Redundancy for View-Decoupled Representation in Multi-View Clustering

カテゴリ：部門大会

論文No：OS4-2-6

グループ名：【C】2025年電気学会電子・情報・システム部門大会

発行日：2025/8/20

タイトル(英語)：Reducing Redundancy for View-Decoupled Representation in Multi-View Clustering

著者名：Ding Yu（富山大学）,Gu Chunzhi（豊橋技術科学大学）,Zhang Chao（富山大学）

著者名(英語)： Yu Ding (University of Toyama),Chunzhi Gu (Toyohashi University of Technology),Chao Zhang (University of Toyama)

キーワード：マルチビュー表現,マルチビュークラスタリング,自己教師あり学習自己教師あり学習,Multi-view representation,Multi-view clustering,Self-supervised learning

要約(日本語)：Multi-view clustering aims to partition multi-view data into semantically meaningful groups. However, many multi-view methods are prone to trivial solutions during training. To address this issue, we propose a deep multi-view clustering framework that avoids representation collapse by guiding intra-view and inter-view correlation matrices toward a target structure. It consists of an intra-view learning module using intra-view augmentations to reduce redundancy, and an inter-view disentanglement module that separates common and unique representations while maintaining semantic consistency. Experiments on standard benchmarks show that our method performs robustly under varying view numbers and yields discriminative, structure-preserving clusters.

本誌掲載ページ：968-972p

原稿種別：英語

PDFファイルサイズ：512Kバイト