Structured Subspace Approach Enhances Clustering Across Incomplete Multi-View Data

A new structured subspace representation approach significantly outperforms the existing Structure Exploration and Missing-view Inference framework in both clustering performance and computational efficiency across multiple incomplete multi-view datasets. Developed by researchers Wanyi Zhang and Kewei Tang and reported in the International Journal of Machine Learning and Cybernetics, the method addresses critical bottlenecks in partitioning multi-view data while remaining competitive with alternative clustering methods.

Incomplete multi-view subspace clustering is designed to partition multi-view data effectively when samples contain missing views. The recently proposed Structure Exploration and Missing-view Inference (SEMI) algorithm attempted to address this task by incorporating structure guidance alongside missing data recovery.

Specifically, SEMI carries out local structure guidance based on graph learning, requiring the nonnegativity of the self-representation. This requirement limits the exploration of data affinities, and the local structure constraints become difficult to satisfy when individual views exhibit significant discrepancies. Additionally, SEMI learns a shared representation that functions equivalently to concatenating all views in a traditional single-view method, which weakens the core advantage of multi-view modeling.

To resolve these limitations, the proposed structured subspace representation imposes soft structure guidance for self-representation learning. By avoiding extra constraints on the representation matrix, the technique effectively extracts data affinity. The framework simultaneously leverages view-specific and unified representations to fully preserve the advantage of multi-view methods when handling view-missing samples.

Published on November 5, 2026, as an original article in Volume 17 (article number 526) of the International Journal of Machine Learning and Cybernetics, the study evaluates the proposed formulation across multiple incomplete multi-view datasets. Across these empirical tests, the experimental results demonstrate that the new methodology significantly outperforms the Structure Exploration and Missing-view Inference approach in both clustering performance and computational efficiency, while concurrently maintaining competitive clustering capability when compared with other alternative methods.

Subject of Research: Incomplete multi-view subspace clustering

Article Title: Structured subspace representation for incomplete multi-view clustering

Article References: Zhang, W., & Tang, K. (2026). Structured subspace representation for incomplete multi-view clustering. International Journal of Machine Learning and Cybernetics, 17(11), Article 526. https://doi.org/10.1007/s13042-026-03368-y

Image Credits: AI Generated

DOI: 10.1007/s13042-026-03368-y

Keywords: incomplete multi-view clustering, subspace clustering, machine learning, data affinity, self-representation, graph learning, computational efficiency, SEMI method, Structured, subspace, representation, incomplete