Dynamic guided metric representation learning for multi-view clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multi-view clustering (MVC) aims to group data from diverse perspectives.
• Effective data representation is crucial for successful MVC.
• Existing methods often struggle to balance intra-view discrimination and inter-view correlation.

Purpose of the Study:

• To propose a novel framework, Dynamic Guided Metric Representation Learning for Multi-View Clustering (DGMRL-MVC).
• To enhance data representation by enforcing class separability within views and consistency across views.
• To achieve improved multi-view clustering performance through a learned latent discriminated embedding space.

Main Methods:

• Utilizing Fisher Discriminant Analysis (FDA) for intra-view class separability.
• Employing Hilbert-Schmidt independence criteria (HSIC) to enhance inter-view consistency.
• Integrating a dynamic routing mechanism and Generalized Canonical Correlation Analysis (GCCA) for a unified representation.

Main Results:

• The DGMRL-MVC framework generates an enhanced, fused representation of multi-view data.
• Experimental results demonstrate significant improvements in multi-view clustering performance.
• The learned latent space effectively captures both discriminative features and cross-view correlations.

Conclusions:

• The proposed DGMRL-MVC framework effectively addresses key challenges in multi-view clustering.
• The method offers a robust approach to learning comprehensive data representations.
• DGMRL-MVC shows strong potential for various multi-view clustering applications.