Malik Magdon-Ismail, Rensselaer Polytechnic Institute – Modeling Infection Rates for Pandemic Control

On Rensselaer Polytechnic Institute Week: Learning to predict how COVID-19 spreads is key.

Malik Magdon-Ismail, professor of computer science, develops one way to do so.

Dr. Magdon-Ismail has been a Professor of Computer Science since 2000. After degrees at Yale and Caltech, Dr. Magdon-Ismail was a research scholar at Caltech before joining Rensselaer as Assistant Professor of Computer Science. His interests are in decision making from data in complex systems, including machine learning, computational finance and social and communication networks. He enjoys poker, bridge, squash, tennis and badminton.

Modeling Infection Rates for Pandemic Control

Modeling techniques can help us understand and prevent the spread of COVID-19 through densely populated urban areas. But in smaller cities and even universities, there’s just not as much information to go on, making it harder to successfully navigate the pandemic. 

I develop specialized machine learning models that harness limited available data to predict the course of the pandemic and the impacts of preventive measures in smaller cities, and at universities like RPI.  In small populations, fewer people get sick, giving you less data to work with. Plot the data on a graph, and you’ll see a lot more variability, and a less-defined pattern. Scientists call that “noisy data,” and standard machine learning algorithms can’t cut through the noise.

Robust machine learning is designed to tune down the noise, find the pattern, and generate accurate predictions.  My robust machine learning models use available data to chart the course of the pandemic in cities and counties throughout all 50 states and in numerous locations worldwide, offering metrics like the rate of infections over time, the rate of asymptomatic infections, and the proportion of a population that must quarantine to maintain a sustainable rate of infection.

We also built COVID Back-to-School, a free online tool for universities and schools.  Users set parameters like the size and geographical origin of the on-campus population, and the average daily number of close interactions for each person during classes and meals, and the model predicts the resulting infection growth. With this tool, we can see the effects of measures like lowering class sizes, requiring masks, and altering the frequency of testing.

In all my work a powerful message makes itself clear: wearing masks, social distancing and proper testing works wonders. If we listen to the data and work together, we can bring this pandemic to heel.