Multi-Attribute Subspace Clustering via Auto-Weighted Tensor Nuclear Norm Minimization.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Existing subspace clustering methods often overlook multi-attribute data characteristics, relying on single self-representations.
• This limitation leads to incomplete data understanding and suboptimal clustering performance in unsupervised learning tasks.

Purpose of the Study:

• To propose a novel Multi-Attribute Subspace Clustering (MASC) model for unsupervised learning.
• To address the limitations of existing methods by exploiting intrinsic multi-attribute features for enhanced data representation.

Main Methods:

• MASC simultaneously learns multiple subspace representations, each corresponding to a specific data attribute.
• Employs an auto-weighted tensor nuclear norm (AWTNN) for low-rank tensor approximation to capture high-order correlations among attribute representations.
• Develops an efficient algorithm to optimize the non-convex MASC model, ensuring convergence.

Main Results:

• The proposed AWTNN effectively captures high-order correlations among multi-attribute representations through adaptive weight splitting.
• The MASC model generates a comprehensive subspace representation by aggregating multi-attribute information.
• Experiments on eight real-world datasets demonstrate MASC's superior performance compared to existing subspace clustering methods.

Conclusions:

• The MASC model offers a significant advancement in subspace clustering by effectively utilizing multi-attribute information.
• The developed AWTNN and optimization algorithm provide a robust framework for non-convex, multi-block model optimization.
• MASC demonstrates superior performance, paving the way for more accurate unsupervised learning on complex, multi-attribute datasets.