.png){kind=icon}
.png){kind=icon}
Return
Block-diagonal regularized multiple kernel k -means clustering for incomplete multi-view data
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2026.134787.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multiple kernel k -means clustering (MKKC) has been successfully applied to incomplete multi-view data (IMVD). However, existing methods suffer from two primary limitations: (1) they fail to assign interpretable and differentiated weights to base kernels when constructing the optimal kernel, and (2) the prevalent multi-step clustering strategy is prone to suboptimal solutions due to error accumulation. To address these issues, we propose a block-diagonal regularized MKKC method for IMVD, termed BD-MKKC-IMV, which incorporates three core innovations. First, a two-stage adaptive kernel weighting strategy assigns weights at both stages: in the first stage, each base kernel is assigned a weight according to its structural similarity to the consensus base kernel; in the second stage, each consensus base kernel is weighted based on its similarity to the optimal kernel. This strategy ensures rational and interpretable weight allocation. Second, a block-diagonal regularization term is imposed on the product of the indicator matrix and its transpose to encourage a clear clustering structure, enabling direct clustering. Third, kernel imputation and clustering assignment are integrated into a unified optimization framework, allowing them to be collaboratively updated and mutually reinforced, thereby preventing cross-stage error propagation. Extensive experiments on six benchmark datasets demonstrate that BD-MKKC-IMV achieves competitive performance compared with eight state-of-the-art methods.