Discovering dynamic models of COVID-19 transmission.

Existing models about the dynamics of COVID-19 transmission often assume the mechanism of virus transmission and the form of the differential equations. These assumptions are hard to verify. Due to the biases of country-level data, it is inaccurate to construct the global dynamic of COVID-19. This research aims to provide a robust data-driven global model of the transmission dynamics. We apply sparse identification of nonlinear dynamics (SINDy) to model the dynamics of COVID-19 global transmission. One advantage is that we can discover the nonlinear dynamics from data without assumptions in the form of the governing equations. To overcome the problem of biased country-level data on the number of reported cases, we propose a robust global model of the dynamics by using maximin aggregation. Real data analysis shows the efficiency of our model.