Tensor-Based Semi-Supervised Multiview Subspace Clustering With Pairwise Constraint Propagation.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Semi-supervised multiview clustering integrates data from multiple sources with limited labels.
• Existing methods underutilize unlabeled data and suffer from unstable outcomes due to two-stage procedures.

Purpose of the Study:

• To propose a unified framework for semi-supervised multiview clustering.
• To enhance the utilization of unlabeled data and improve clustering stability.

Main Methods:

• Developed a tensor-based semi-supervised multiview subspace clustering (TSMSC) method.
• Integrated multiview subspace clustering with pairwise constraint propagation using a unified tensor framework.
• Decomposed subspace representations into consensus and private parts, and employed low-rank matrix completion for constraint propagation.

Main Results:

• The proposed TSMSC method demonstrated superior performance across eight real-world datasets.
• Achieved more stable and accurate clustering outcomes compared to state-of-the-art methods.
• Effectively leveraged both labeled and unlabeled data for improved clustering.

Conclusions:

• The unified tensor-based framework offers a significant advancement in semi-supervised multiview clustering.
• The method effectively addresses the limitations of existing approaches by joint optimization.
• TSMSC provides a robust and efficient solution for complex multiview data analysis.