Structural deep multi-view clustering with integrated abstraction and detail.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Deep multi-view clustering leverages complementary information from diverse data sources.
• Existing methods often neglect structural information and are sensitive to data quality, limiting clustering performance.
• There is a need for robust multi-view clustering that integrates both abstract and detailed feature representations.

Purpose of the Study:

• To propose Structural deep Multi-View Clustering (SMVC), a novel framework that integrates abstraction and detail for improved clustering.
• To address limitations of existing methods by incorporating structural information and enhancing robustness to view quality.
• To jointly optimize cluster assignments and feature embeddings within a unified model.

Main Methods:

• Feature extraction from individual views using multi-layer perceptrons, followed by concatenation for global features.
• Construction of a global target distribution to guide soft cluster assignments across views.
• Instance-level contrastive learning using high-order adjacency matrices to mine underlying details and reduce redundancy, mimicking graph attention network effects.

Main Results:

• The proposed SMVC framework effectively integrates high-level abstract features with low-level detailed features.
• The model demonstrates superior performance in jointly optimizing cluster assignments and feature embeddings.
• Extensive experiments on four benchmark datasets show SMVC consistently outperforms state-of-the-art methods.

Conclusions:

• SMVC offers a significant advancement in deep multi-view clustering by incorporating structural information and a dual-level feature integration strategy.
• The model's ability to leverage both abstraction and detail leads to enhanced clustering accuracy and robustness.
• SMVC represents a promising direction for future research in multi-view representation learning and clustering.