Fast Incomplete Multi-View Clustering With View-Independent Anchors

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Multi-view clustering (MVC) methods leverage information across multiple data views for improved performance.
• Incomplete multi-view clustering (IMC) addresses datasets where instances have missing information across views.
• Existing fast IMC methods often overlook view-specific information when processing large-scale incomplete data.

Purpose of the Study:

• To propose a novel fast incomplete multi-view clustering method named FIMVC-VIA.
• To address the limitations of existing IMC methods that ignore view-specific information.
• To develop a scalable solution for large-scale incomplete multi-view clustering tasks.

Main Methods:

• Proposed FIMVC-VIA learns individual anchors based on the distribution diversity of each incomplete view.
• A unified anchor graph is constructed using the principle of consistent clustering structure.
• The method constructs an anchor graph instead of a full pairwise graph to reduce complexity.

Main Results:

• FIMVC-VIA achieves linear time and space complexity concerning the number of samples, enabling efficient processing of large-scale data.
• Experiments demonstrate improved effectiveness and complexity compared to other IMC methods across various missing rates.
• The proposed method effectively handles partially available information in multi-view datasets.

Conclusions:

• FIMVC-VIA offers an efficient and effective solution for incomplete multi-view clustering, particularly for large-scale datasets.
• The use of view-independent anchors and anchor graph construction significantly enhances scalability and performance.
• The method's ability to exploit both consistent and complementary information across views, even with missing data, is a key advantage.