Stochastic Graphlet Embedding.

Area of Science:

• Computer Science
• Machine Learning
• Pattern Recognition

Background:

• Graph-based methods are effective for machine learning and pattern classification.
• Non-vectorial graph data requires embedding into vector spaces for standard algorithms.
• Existing embedding methods must be resilient to variations and highly discriminant.

Purpose of the Study:

• To propose a novel high-order stochastic graphlet embedding method.
• To map graphs into vector spaces for improved machine learning performance.
• To address the challenge of representing complex graph structures.

Main Methods:

• Developed a stochastic search procedure for efficient graph parsing.
• Introduced high-order graphlet extraction and sampling.
• Utilized hash functions for efficient isomorphic set assignment with low collision probability.
• Integrated graphlet distributions with maximum margin classifiers.

Main Results:

• The proposed method successfully maps graphs into vector spaces.
• High-order graphlets effectively model local primitives and their interactions.
• Graphlet-based representations positively impact pattern comparison and recognition.
• Extensive experiments on benchmark databases corroborate performance improvements.

Conclusions:

• The novel high-order stochastic graphlet embedding offers a robust approach for graph data.
• This method enhances the discriminative power of graph representations in machine learning.
• The technique shows significant promise for pattern recognition tasks.