Incomplete Data Meets Uncoupled Case: A Challenging Task of Multiview Clustering

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Existing Incomplete Multiview Clustering (IMC) methods excel with complementary information but struggle with simultaneously uncoupled and incomplete data.
• Uncoupled incomplete data presents challenges in cross-view correlation, hindering effective complementary information exploration and leading to suboptimal clustering.
• Hyperparameters in current IMC methods limit their practical applicability.

Purpose of the Study:

• To propose a novel Uncoupled Incomplete Multiview Clustering (UIMC) method capable of handling both uncoupled and incomplete multiview data.
• To develop a joint framework for feature inference and recoupling to overcome limitations of existing IMC techniques.
• To enhance the generalization ability and clustering performance of IMC methods in real-world scenarios.

Main Methods:

• Developed a joint framework for feature inferring and recoupling to address uncoupled incomplete multiview data.
• Employed tensor singular value decomposition (t-SVD)-based tensor nuclear norm (TNN) to explore high-order correlations in recoupled and inferred self-representation matrices.
• Implemented an exploratory approach for updating all hyperparameters within the UIMC method.

Main Results:

• The proposed UIMC method demonstrates superior performance in handling uncoupled incomplete multiview data.
• Experiments on six real-world datasets validate the effectiveness of UIMC compared to state-of-the-art methods.
• The joint framework and t-SVD-based TNN effectively capture complex correlations in challenging data.

Conclusions:

• The novel UIMC method successfully addresses the simultaneous challenge of uncoupled and incomplete multiview data.
• UIMC offers improved clustering performance and generalization capabilities for real-world applications.
• The exploratory hyperparameter updating mechanism enhances the method's practicality and robustness.