Accelerated Partially Shared Dictionary Learning With Differentiable Scale-Invariant Sparsity for Multi-View...

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Multiview dictionary learning (DL) is crucial for feature learning in multiview clustering.
• Existing algorithms struggle to fully leverage consistent and complementary information simultaneously due to inter-view gaps.

Purpose of the Study:

• To propose an efficient multiview DL algorithm for improved multiview clustering.
• To address the limitations of existing methods in utilizing both consistent and complementary information.

Main Methods:

• A partially shared DL model with a flexible ratio of shared sparse coefficients is employed.
• A differentiable scale-invariant function serves as a sparsity regularizer, approximating the l0 norm.
• Optimization is handled using the proximal splitting method with extrapolation technology.

Main Results:

• The proposed algorithm effectively recovers synthetic dictionaries with reasonable convergence times.
• Experiments on six real-world datasets show robustness to regularizer parameter changes.
• An optimal coefficient sharing ratio enhances clustering by exploiting both consistent and complementary information.

Conclusions:

• The developed algorithm achieves superior multiview clustering performance compared to existing methods.
• The approach demonstrates faster convergence rates in multiview clustering tasks.
• The flexible coefficient sharing ratio is key to effectively utilizing multiview data characteristics.