A Novel Low-Rank Embedded Latent Multi-View Subspace Clustering Approach

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Noises and outliers degrade prediction performance in data processing.
• Multi-view learning integrates information from heterogeneous modalities but faces challenges like view inconsistencies and misalignment.
• Existing methods often rely on explicit-view clustering and strict alignment assumptions.

Purpose of the Study:

• To develop a robust and efficient multi-view feature fusion method.
• To address challenges posed by noise, outliers, and misalignment in multi-view data.
• To improve the effectiveness of multi-view learning in practical data processing.

Main Methods:

• Proposed a latent multi-view representation learning model based on low-rank embedding.
• Utilized low-rank constraints to create a unified latent subspace representation.
• Incorporated an adaptive noise suppression mechanism.
• Employed the Augmented Lagrangian Multiplier Alternating Direction Minimization (ALM-ADM) framework for optimization.

Main Results:

• The proposed model effectively uncovers latent consistency structures across different views.
• Demonstrated enhanced robustness against outliers and noise interference.
• Achieved efficient multi-view feature fusion.
• Outperformed state-of-the-art methods in clustering performance and robustness on benchmark datasets.

Conclusions:

• The latent multi-view representation learning model offers a robust and efficient solution for multi-view feature fusion.
• The method successfully mitigates the impact of noise and misalignment.
• The approach shows significant improvements in clustering tasks and overall data processing robustness.