Enhanced Residual Tensor Norm Minimization for Multiview Subspace Clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Low-rank tensor constraints are crucial for multiview subspace clustering (MSC).
• Existing methods face challenges with tensor rank surrogates and tensor rotation operations.
• These operations are critical for enhancing performance in multiview settings.

Purpose of the Study:

• To address limitations in current low-rank tensor constraint methods for MSC.
• To introduce a novel approach, the Enhanced Residual Tensor Norm (ERTN), for improved multiview clustering.
• To enhance the exploitation of structural information in multiview data.

Main Methods:

• Developed ERTN using a novel surrogate for tensor rank based on residual learning of singular values.
• Applied tensor-singular value decomposition (t-SVD) on three modes of the constructed tensor.
• Utilized an augmented Lagrangian multiplier-based algorithm for optimization with convergence guarantees.

Main Results:

• ERTN facilitates better exploitation of structural information in multiview data.
• t-SVD on three modes generalizes tensor rotation, exploring intra- and inter-view information comprehensively.
• Experiments on real-world datasets demonstrate the effectiveness of ERTN-MSC.

Conclusions:

• ERTN-MSC offers a competitive and effective approach to multiview subspace clustering.
• The proposed method overcomes key challenges in existing low-rank tensor constraint techniques.
• ERTN-MSC shows significant potential for advancing multiview data analysis and clustering.