Anchor point segmentation based multi-view clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Existing bipartite graph methods for multi-view clustering use anchor graphs and optimization techniques.
• These methods maintain linear complexity but under-explore the geometric relationship between anchors and raw data centroids.
• This under-exploration limits potential improvements in algorithm efficiency.

Purpose of the Study:

• To propose a novel multi-view clustering approach, anchor point segmentation based multi-view clustering (APS-MVC).
• To efficiently learn clustering centroids by utilizing the geometric relationship between anchors and raw data.
• To address the out-of-sample issue in multi-view clustering.

Main Methods:

• APS-MVC assigns data points to anchor points, then to centroids, modeled as a two-step Markov chain transition.
• Optimal centroids and anchor soft partitions are learned simultaneously by encoding anchor graph structure.
• The approach exhibits square complexity concerning the number of anchors.

Main Results:

• Experimental results on six benchmark datasets demonstrate the effectiveness of APS-MVC.
• The method efficiently solves the optimization problem.
• APS-MVC effectively addresses the out-of-sample challenge.

Conclusions:

• APS-MVC offers an efficient and effective solution for multi-view clustering.
• The novel approach improves upon existing methods by leveraging anchor-data geometric relationships.
• The proposed method shows strong performance across various datasets.