.png){kind=icon}
.png){kind=icon}
Return
Deep Semantic Prototype Alignment for Incomplete Multi-View Clustering
.png){kind=icon}
PRE
AI
DOI：10.1109/LSP.2025.3618773.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-view clustering enhances clustering performance by integrating information from multiple sources, but partial view missingness poses significant challenges for Incomplete Multi-view Clustering (IMC). Existing matrix factorization (MF) based methods achieve promising results by learning to implicitly impute missing data through a shared latent space. However, the current single-layer MF framework cannot fully utilize useful data representations, making it prone to imputation errors and thereby affecting the quality of view alignment. To address this issue, this letter proposes a Deep Semantic Prototype Alignment Incomplete Multi-view Clustering method (DSPA-IMC). First, hierarchical local representations are learned directly from the observed data of each view through deep matrix factorization, thereby reducing the error caused by speculative imputation. Second, we propose a prototype-driven semantic alignment mechanism that projects deep local representations into a shared semantic space defined by orthogonal prototypes, effectively achieving cross-view semantic unification and guiding the handling of missing data. Experimental results demonstrate that DSPA-IMC outperforms eleven existing state-of-the-art methods across various missing rates.
Keywords:
Incomplete multi-view clustering
deep matrix factorization
prototype-driven alignment
Journal
I
IF:
3.9
Papers:
591
Citations:
0