Cross-View Approximation on Grassmann Manifold for Multiview Clustering.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Existing multiview clustering methods struggle with simultaneous learning from graph and feature spaces.
• Inconsistent clustering structures and the need for postprocessing are common limitations.

Purpose of the Study:

• To propose a Cross-view Approximation on Grassmann Manifold (CAGM) model.
• To address inconsistencies in multiview adjacency, feature matrices, and cross-view combinations.
• To achieve a consistent clustering structure without postprocessing.

Main Methods:

• Utilizes a ratio-formed objective function for parameter-free bidirectional fusion.
• Incorporates a paired encoding mechanism for low-dimensional, orthogonal cross-view embeddings.
• Employs approximation of measurable subspaces on the Grassmann manifold for direct indicator matrix acquisition.

Main Results:

• Demonstrates effective handling of inconsistencies within multiview data.
• Achieves direct acquisition of the indicator matrix.
• Validated through comprehensive experiments on four real-world datasets.

Conclusions:

• The proposed CAGM model effectively improves multiview clustering performance.
• CAGM offers a unified framework that integrates graph and feature learning.
• The method eliminates the need for postprocessing steps in multiview clustering.