Parameter-Free and Scalable Incomplete Multiview Clustering With Prototype Graph.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Multiview clustering (MVC) effectively groups data but struggles with incomplete data views.
• Existing incomplete multiview clustering (IMVC) methods face scalability issues due to high time and space complexity.
• Current IMVC approaches often require extensive hyper-parameter tuning, limiting practical application.

Purpose of the Study:

• To propose a novel, scalable, and parameter-free framework for incomplete multiview clustering.
• To address the limitations of existing IMVC methods in terms of computational complexity and hyper-parameter dependency.
• To leverage prototype-based learning for enhanced IMVC performance.

Main Methods:

• Developed PSIMVC-PG, a parameter-free and scalable incomplete multiview clustering framework.
• Constructed an incomplete prototype graph to capture instance-prototype relationships, differing from pairwise graph methods.
• Integrated prototype graph construction directly, eliminating the need for hyper-parameter pre-searching.

Main Results:

• PSIMVC-PG demonstrates significant advantages over existing IMVC methods in extensive experiments.
• The framework achieves superior performance on various incomplete multiview clustering tasks.
• The proposed method effectively handles incomplete data views with improved scalability.

Conclusions:

• PSIMVC-PG offers a practical and efficient solution for large-scale incomplete multiview clustering.
• The parameter-free nature and prototype graph approach enhance usability and performance.
• This work advances the field of IMVC by providing a scalable and robust framework.