RAPID: Networked Data-Driven Modelling of the COVID-19 Outbreak with a Performativity-Aware Calibration Learning Algorithm

Mathematical and Physical Sciences - This project will develop and deploy a data-driven mathematical modeling framework for predicting the spread of COVID-19 at regional levels and for informing potential mitigation efforts. The models will also provide a means to test the impact of social distancing and mobility reduction on the future course of the pandemic. The proposed modeling framework relies on a two-component structure that does not require prior knowledge of the epidemiological characteristics of the disease. This approach is especially useful during the initial stages of an emerging outbreak, where little is known and validated about the contagion. Moreover, this project will bring a novel perspective on the mathematical modeling of disease spread, which will complement other ongoing efforts and provide access to diverse models critical to decision-making under uncertainty.

This project builds upon a data-driven mathematical modeling approach leveraging a surprisingly simple behavior examined in epidemiological data sets and models that allows forecasts for case counts with no parameter estimations. The first thrust is to integrate data-driven modeling into explicit network interactions in order to investigate spatial aspects of COVID-19 outbreak propagation. The second thrust of the project is to implement a calibration layer that takes into account mitigation efforts. The rationale for this approach is that, in a constantly evolving environment, epidemiological predictions are difficult to make due to the performativity effect, whereby model predictions affect social behavior and mitigation efforts, which in turn alters the spread of the outbreak predicted by the mathematical models. From a conceptual point of view, this project will address performativity in the context of epidemiological modeling. At the practical level, it will develop a general calibration module that will learn how to incorporate reactions to predictions into epidemiological forecasts. By design, this ?performativity-aware? calibration module will be independent of any specific epidemic model; hence, once developed, it will be possible to be integrated into other existing predictive models.

This award is being funded by the CARES supplemental funds allocated to MPS.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.