Multiview Spectral Clustering Based on Consensus Neighbor Strategy.

Area of Science:

• Data Mining
• Machine Learning
• Computer Vision

Background:

• Multiview spectral clustering is valuable for spatial learning but limited by constrained consensus adjacency matrix search spaces.
• Existing methods restrict the optimal consensus adjacency matrix within the span of individual view adjacency matrices, hindering performance.

Purpose of the Study:

• To propose a novel and convex consensus neighbor strategy for learning the optimal consensus adjacency matrix in multiview spectral clustering.
• To overcome the limitations of existing methods by expanding the feasible domain for discovering the optimal consensus adjacency matrix.

Main Methods:

• Developed a consensus neighbor strategy to capture consensus local structure across all views for constructing the optimal consensus adjacency matrix.
• Introduced a correlation measuring matrix to prevent trivial solutions.
• Designed an efficient iterative algorithm leveraging the model's convex nature for guaranteed convergence to a global optimum.

Main Results:

• The proposed algorithm demonstrated superior consensus representation learning capabilities compared to state-of-the-art methods.
• Experimental validation on 16 multiview datasets confirmed the effectiveness of the consensus neighbor strategy.
• The method successfully expanded the search space for the optimal consensus adjacency matrix.

Conclusions:

• The consensus neighbor strategy offers a robust and effective approach to multiview spectral clustering.
• This work advances the field of representation learning by enabling broader exploration of consensus adjacency matrices.
• The developed algorithm provides a reliable and efficient solution for complex data mining tasks.