Machine learning assisted network classification from symbolic time-series

Area of Science:

• Complex systems analysis
• Network science
• Machine learning applications

Background:

• Machine learning excels at predicting complex system behaviors.
• Analyzing large-scale networks often requires extensive data.
• Classifying network structures is crucial for understanding system dynamics.

Purpose of the Study:

• To develop a simplified deep learning method for classifying network structures.
• To determine if limited time-series data can accurately identify network types.
• To demonstrate the method's efficacy on coupled Kuramoto oscillators and susceptible-infectious-susceptible models.

Main Methods:

• Applied deep learning to limited time-series data from a few nodes.
• Focused on systems in a partially synchronized state.
• Utilized binary time-series information for classification.

Main Results:

• Accurate classification of underlying network structures was achieved using minimal node data.
• Binary time-series information proved as effective as actual time-series data.
• The method successfully classified network structures in coupled Kuramoto oscillators and susceptible-infectious-susceptible models.

Conclusions:

• Deep learning with limited node data offers a simple yet accurate method for network structure classification.
• Partially synchronized states are key for effective analysis.
• This approach simplifies the prediction of large-scale network classifications.