Pandemic Modeling

Goal

Accurately forecast COVID-19 cases, admissions, and deaths at the federal, state, and local levels as well as identify and project the impact of social distancing policies, weather, and other factors on the disease’s ability to spread using state of the art machine learning techniques in conjunction with specially curated data from our partner team task geography.

Our Difference

There are a lot of different groups attempting to forecast COVID-19 and its spread. Our approach, however, is fundamentally different in that rather than employ simple mathematical models we seek to utilize some of the most big and complex neural network architectures. While this approach might seem unintuitive, these large neural network for time series forecasting architectures enable the model to learn on many different covariates and predict several outputs (admissions, deaths, new cases) at once rather than requiring separate models. Of course, this means that we also require a LOT of data. However, thanks to our partnership with task-geo, we have high-quality temporal data on COVID-19. Additionally, we plan on extensively leveraging transfer learning and other time series data augmentation methods. Altogether we aim to construct an effective template for using deep learning to forecast COVID-19 that can also be applied to future epidemics with limited data (or any other data constrained forecasting scenario).

Team

Isaac McKillen-Godfried Team Leader Email Aradhana Data Scientist Mike Honey Data Visualization Specialist Iason Konstantinidis Kevin Li Data Visualization Maggie Data Scientist Malavika Suresh Data Scientist Wendy Mak

Recruiting

We are actively looking for skilled epidemiologists, virologists, as well as additional data scientists willing to donate their time for this important endeavor. Please do not hesitate reach out to the team lead if at all interested.