Exploring Spatio-temporal Dynamics of Cellular Automata for Pattern Recognition in Networks

Area of Science:

• Network science
• Complex systems analysis
• Computational biology

Background:

• Network science offers integrative approaches to complex systems.
• Network modeling is increasingly used for emergent phenomena in real-world applications.
• Network characterization is crucial for understanding topological properties and network models.

Purpose of the Study:

• Introduce the Life-Like Network Automata (LLNA) method for network pattern recognition.
• Utilize network topology as a Cellular Automata (CA) tessellation for feature extraction.
• Evaluate LLNA's performance on synthetic and real-world network data.

Main Methods:

• Life-Like Network Automata (LLNA) models network topology as Cellular Automata (CA).
• CA dynamics generate spatio-temporal patterns for feature vector extraction.
• LLNA was tested on metabolic, social, and biological networks.

Main Results:

• LLNA demonstrated superior performance compared to structural measurements in real-world applications.
• Classification rate improvements of up to 23% were observed in identifying organisms via metabolic networks.
• Significant improvements were also noted in identifying online social networks (4%) and classifying stomata patterns (7%).

Conclusions:

• The proposed LLNA method is effective for pattern recognition in networks.
• LLNA shows potential for broad applicability in diverse network analysis tasks.
• LLNA offers a robust approach to network characterization and feature extraction.