Decomposing graphs to balance virus spreading and efficiency using Graph Neural Networks

Abstract

Maintaining a high system performance while imposing virus preventing measures is a challenging task. We introduce a Deep Epidemic Efficeincy Network (DEEN) which balances the two opposing goals via graph partition. Our model optimises graph efficiency while meeting increasing levels of the epidemic threshold. We introduced our method to the ride-pooling service in New York City. By dividing 150 New York taxi travellers into four groups, our method increases the epidemic threshold by more than twofold at the cost of reducing utility only by 13%. We validate our model against other real-world examples: cross-region economic exchange in Poland and information sharing in a peer-to-peer network.