Learning epidemic threshold in complex networks by Convolutional Neural Network.

Area of Science:

• Complex Networks
• Epidemiology
• Machine Learning
• Deep Learning

Background:

• Deep learning models excel in Euclidean space but struggle with complex network structures.
• Existing models cannot effectively integrate both structural and dynamical information of real-world networks.
• Understanding epidemic thresholds is crucial for public health and network management.

Purpose of the Study:

• To develop a novel framework for learning epidemic thresholds in complex networks.
• To effectively combine structural and dynamical information for improved prediction accuracy.
• To create a robust and universally applicable machine learning model for arbitrary network topologies.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for their proven performance in Euclidean space learning.
• Employed graph representation learning for dimensionality reduction of network data.
• Converted network data into an image-like structure and merged nodal dynamics using multichannel images.

Main Results:

• The proposed framework accurately identifies epidemic outbreak thresholds using a 'confusion scheme'.
• Demonstrated strong performance on both synthetic and empirical network datasets.
• The model successfully integrates structural and dynamical information for enhanced learning.

Conclusions:

• The developed end-to-end machine learning framework is robust and effective for complex networks.
• This approach offers a universally applicable solution for predicting epidemic thresholds.
• The method provides a significant advancement in applying deep learning to network dynamics.