Supervised Learning Epidemic Threshold of SIR Model in Complex Networks

Classifying phase transitions of epidemic outbreaks in different kinds of complex networks is a central problem in network dynamics research. Deep learning methods can be used to identify phases and phase transitions in complex networks via supervised machine learning. However, most studies recently published focus on dynamical information of a single node. As a matter of fact, structural features in complex networks also play a significant role in dynamics progress. In this paper, we propose a novel deep learning framework to combine the structural and dynamical information into an image with multiple channels. Then, convolutional neural network (CNN) is used to find phase transition depending on supervised learning labeled image data. By training on regular random network data and scale-free network data, we show our machine learning framework can learn the epidemic threshold of SIR model in a high accuracy and robustness. What’s more, complex networks with arbitrary topology and size and real networks can be used universally.