Tensorized anchor alignment for incomplete multi-view clustering.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Incomplete Multi-View Clustering (IMVC) aims to leverage consensus and complementary information from datasets with missing views.
• Existing IMVC methods often suffer from high computational complexity, anchor misalignment, and failure to capture high-order correlations.
• Addressing these limitations is crucial for developing more effective and efficient clustering techniques.

Purpose of the Study:

• To introduce a novel framework, Tensorized Anchor Alignment for Incomplete Multi-view Clustering (TAA-IMC), to overcome the limitations of current IMVC methods.
• To enhance the efficiency and accuracy of clustering incomplete multi-view data.
• To effectively handle anchor misalignment and extract high-order correlations among multiple views.

Main Methods:

• Constructing view-specific anchor graphs to reduce computational complexity and preserve data diversity.
• Employing a binary alignment matrix to ensure accurate anchor correspondence across different views, mitigating misalignment.
• Integrating aligned anchor graphs into a low-rank tensor representation to capture high-order correlations, utilizing an alternating update method for solution.

Main Results:

• The proposed TAA-IMC framework demonstrates significant efficiency in terms of memory and time complexity.
• Extensive experiments on seven benchmark datasets show TAA-IMC outperforms existing state-of-the-art methods.
• The method effectively addresses anchor misalignment and extracts valuable high-order correlation information.

Conclusions:

• TAA-IMC offers an efficient and superior solution for incomplete multi-view clustering problems.
• The tensor-based approach effectively captures complex relationships within multi-view data.
• The framework provides a robust method for handling missing data and improving clustering accuracy.