Adaptive Anchor-Guided Representation Learning for Efficient Multi-View Subspace Clustering.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Multi-view Subspace Clustering (MVSC) aggregates data from multiple sources for improved clustering.
• Existing anchor-based MVSC methods face challenges in capturing both consistency and complementary information simultaneously.
• High computational complexity, particularly from Singular Value Decomposition (SVD), limits the scalability of current MVSC techniques.

Purpose of the Study:

• To propose an Adaptive Anchor-guided Representation Learning for Efficient Multi-view Subspace Clustering (A2RL-EMVSC) framework.
• To enhance MVSC performance and scalability by addressing limitations in existing methods.
• To develop a method that simultaneously exploits consistency and complementary information from multiple views.

Main Methods:

• The A2RL-EMVSC framework integrates consensus anchors learning, anchor-guided representation learning, and matrix factorization.
• It learns view-specific anchor representation matrices guided by consensus anchors.
• Matrix decomposition is applied to view-specific matrices for efficient clustering result generation.

Main Results:

• The proposed method effectively captures both consistency and complementary information across multiple views.
• Clustering results are obtained with linear time complexity, significantly improving scalability.
• Extensive experiments on ten datasets demonstrate superior clustering effectiveness compared to state-of-the-art methods.

Conclusions:

• A2RL-EMVSC offers an effective and efficient solution for Multi-view Subspace Clustering.
• The framework successfully balances capturing data consistency and complementary information.
• The proposed method represents a significant advancement in scalable and high-performance MVSC.