Spectral Embedding Representation Based on Random Anchor Graph Aggregation.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Anchor-based strategies accelerate spectral clustering but depend on anchor quality.
• Random sampling is efficient but single samples may not capture data topology.
• Existing methods struggle with accurate topological representation and performance on large datasets.

Purpose of the Study:

• To propose a novel spectral embedding representation model, Random Anchor Graph Aggregation (RAGA).
• To enhance sample representation capability through an aggregated anchor graph.
• To improve clustering performance and handle large-scale data scenarios.

Main Methods:

• Multiple random samplings to approximate original data distribution.
• Adaptive weighted learning for constructing an aggregated anchor graph.
• Joint spectral embedding and spectral rotation learning framework.

Main Results:

• The aggregated anchor graph more precisely portrays the topological structure of original samples.
• RAGA reduces model learning error accumulation compared to traditional two-stage frameworks.
• Experimental results demonstrate RAGA outperforms state-of-the-art anchor graph-based clustering methods.

Conclusions:

• RAGA maintains the speed advantage of random sampling while achieving high-quality anchor graph aggregation.
• The proposed method effectively handles large-scale data scenarios.
• RAGA offers superior clustering performance compared to existing anchor graph-based methods.