.png){kind=icon}
.png){kind=icon}
Return
Dynamic anchor-based deep multi-view subspace clustering network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.130290.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep multi-view subspace clustering methods have attracted increasing attention due to their excellent performance. However, existing methods based on self-representation property are limited in analyzing large-scale data because of the high time and space complexity. To address this problem, this paper proposes a dynamic anchor-based deep multi-view subspace clustering network (DADMVSC) by exploring the representation relationships between latent and anchor features of multi-view data. DADMVSC integrates view-specific latent feature extraction, dynamic anchor learning, and weighted anchor-sample representation relationship learning into a unified end-to-end network. Specifically, a shared dynamic anchor learning network cooperatively working with the deep auto-encoder is designed to extract the view-specific latent features and anchor features autonomously for multiple views. Based on the anchor and latent features, the fully connected layers are inserted between the encoder and dynamic anchor learning network to learn the anchor-sample representation matrices of multi-view data. We construct the anchor-sample representation tensor and employ a channel attention network to model their higher-order relations as discriminative weights for view-specific representation matrices. Subsequently, the weighted anchor-sample representation matrices are fused by concatenation operation to obtain the multi-view consensus representation. Finally, fast spectral clustering is performed on the consensus representation to achieve efficient clustering. The effectiveness of DADMVSC is demonstrated on large-scale multi-view datasets, and the efficiency is verified by algorithm complexity theory analysis and running time comparison experiments. DADMVSC provides a simple and scalable but effective network framework for deep multi-view subspace clustering analysis on large-scale data.
Keywords:
Deep multi-view subspace clustering
Dynamic anchor learning
Anchor-sample representation learning
Large-scale data
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W