Bi-Nuclear Tensor Schatten-p Norm Minimization for Multi-View Subspace Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multi-view subspace clustering integrates information from diverse data sources.
• Low-rank representation (LRR) is a benchmark but suffers from performance limitations and high computational costs.
• Existing LRR methods often use biased nuclear norms and involve computationally expensive Singular Value Decomposition (SVD).

Purpose of the Study:

• To address the limitations of existing low-rank representation methods in multi-view subspace clustering.
• To propose a novel method that enhances clustering performance and computational efficiency.
• To explore high-order correlations and subspace structures within multi-view features.

Main Methods:

• Proposed Bi-nuclear tensor Schatten-p norm minimization (BTMSC) for multi-view subspace clustering.
• Constructed a third-order tensor to capture high-order correlations and subspace structures.
• Utilized Bi-Nuclear Quasi-Norm (BiN) factorization for efficient tensor decomposition.
• Developed an alternating optimization algorithm to solve the BTMSC model.

Main Results:

• BTMSC demonstrated superior performance compared to state-of-the-art methods across ten text and image datasets.
• The proposed method effectively captures low-rank properties in both intra-view and inter-view dimensions.
• BiN factorization significantly improved computational efficiency by factorizing the tensor into smaller components.

Conclusions:

• BTMSC offers a significant advancement in multi-view subspace clustering.
• The method overcomes the limitations of traditional LRR approaches, providing better accuracy and efficiency.
• BTMSC's tensor-based approach effectively leverages multi-view data for improved clustering outcomes.