Tensorized and Compressed Multi-View Subspace Clustering via Structured Constraint.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Traditional multi-view learning often overlooks view consistency, leading to issues with noisy or abnormal data.
• High-dimensional and large-scale datasets pose significant computational challenges for existing methods.

Purpose of the Study:

• To propose a novel multi-view compressed subspace learning method.
• To unify multi-view learning and clustering within a single framework.
• To enhance robustness against noisy data and improve computational efficiency.

Main Methods:

• Utilizes partial samples to construct small dictionaries for each view, reducing redundancy and cost.
• Imposes a low-rank tensor constraint to capture consistency and differences among views.
• Incorporates an auto-weighted mechanism for optimal representation learning.
• Employs a bipartite graph for direct clustering without post-processing.

Main Results:

• Demonstrates superior efficacy and efficiency on synthetic and real-world benchmark datasets.
• Shows particular effectiveness in handling views with noise or outliers.
• Achieves direct clustering results through a structured bipartite graph approach.

Conclusions:

• The proposed method offers an effective and efficient solution for multi-view compressed subspace learning.
• The unified framework successfully integrates clustering and learning, improving robustness.
• The approach is well-suited for modern high-dimensional and large-scale datasets, especially those with data imperfections.