Fast Multi-View Clustering via Nonnegative and Orthogonal Factorization.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Large-scale data clustering presents significant challenges, particularly with multi-view data from diverse sources.
• Existing accelerated multi-view methods often suffer from high computational complexity, limiting their application in high-efficiency scenarios.

Purpose of the Study:

• To propose a fast multi-view clustering model (FMCNOF) to overcome the high computational complexity of existing methods for large-scale data.
• To develop an efficient optimization algorithm for the proposed FMCNOF model.

Main Methods:

• Introduced a novel fast multi-view clustering model (FMCNOF) using nonnegative and orthogonal factorization.
• Constrained one factor matrix as a cluster indicator matrix for direct label assignment, eliminating post-processing.
• Utilized the F-norm for easier optimization and developed an efficient algorithm dividing the problem into smaller subproblems.

Main Results:

• The FMCNOF model significantly improves clustering efficiency with nearly O(n) computational complexity.
• The proposed optimization algorithm reduces matrix multiplications compared to traditional methods.
• Experimental results on benchmark datasets demonstrate substantial efficiency gains while maintaining acceptable performance.

Conclusions:

• The FMCNOF model and its optimization algorithm offer a highly efficient solution for large-scale multi-view clustering.
• This approach effectively addresses the limitations of existing methods in terms of speed and computational cost.
• The method provides a viable option for scenarios demanding high-efficiency clustering of complex, multi-view datasets.