Augmenting Epidemic Models with Graph Neural Networks

Abstract

Conventional epidemic models are limited in their ability to capture the dynamics of real world epidemics in a sense that they either place restrictions on the models such as their topology and contact process for mathematical tractability or focus only on the average global behavior, which lacks details for further analysis. We propose a novel modeling approach that augments the conventional epidemic models using Graph Neural Networks to improve their expressive power while preserving useful mathematical structures. Simulation results show that our proposed model can predict spread times in both node-level and network-wide perspectives with high accuracy having median relative errors below 15% for a wide range of scenarios.