Anchor Graph Network for Incomplete Multiview Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Incomplete multiview clustering (IMVC) is a challenging task.
• Existing IMVC methods often overlook sample pair correlations and are computationally expensive.
• Refinement of bipartite graph structures is frequently neglected in current approaches.

Purpose of the Study:

• To address the limitations of existing IMVC methods.
• To propose a novel anchor graph network for efficient and effective IMVC.
• To enhance the handling of large-scale incomplete data clustering.

Main Methods:

• A generative model is employed to construct bipartite graphs, capturing latent global structure distributions.
• Graph Convolutional Networks (GCNs) utilize these bipartite graphs for learning structural embeddings.
• An adaptive learning strategy is incorporated for robust bipartite graph construction.

Main Results:

• The proposed method significantly reduces computational complexity, enabling scalability to large datasets.
• Bipartite graphs are used to guide the GCN learning process, unlike previous methods.
• Experimental results show comparable or superior performance against state-of-the-art IMVC techniques.

Conclusions:

• The novel anchor graph network offers an efficient and effective solution for IMVC.
• The integration of bipartite graphs and GCNs improves the handling of global structures and reduces computational cost.
• The adaptive learning strategy enhances the robustness of bipartite graph construction for IMVC.