Facilitating Machine Learning Research to Inform Coronavirus Response

Original article can be found here (source): Artificial Intelligence on Medium

The machine learning community should actively engage in these discussions and contribute possible solutions to actionable problems.

One interesting direction could be to identify the effect that different mitigation and suppression strategies have in terms of benefits and costs. “Benefits” in this case would correspond to reductions in the effective reproduction number R, potential lives saved and long-term socio-economic benefits, while “costs” could reflect the resulting burden on the healthcare system, short-term economic consequences, and possible long-term economic restructuring.

Many of the recent epidemiological predictions and analyses are performed for the US as a whole. However, identifying relationships between “benefits” and “costs” will likely require a much higher granularity of analysis. This is because highly localized contextual factors, such as population density, demographics, hospital capacity, or primary means of transportation, will affect critical parameters for computational epidemiological modeling, including the effective reproduction number R. The COVID-19 outbreak on the Diamond Princess cruise ship hints at this circumstance.

The same argument holds for predicting the “cost” of interventions, since certain industries are much more strongly affected than others. This suggests that some areas with a susceptible demographic or socioeconomic composition might be much more severely impacted by aggressive interventions that target disease suppression.

Thinking about this problem raises several underlying questions:

  • Is there a relationship between demographic, socioeconomic, or other factors and the spread of the disease? If yes, what are those factors and how can that relationship be understood? This is one of the ideas behind the recent COVID-19 Forecasting Challenge.
  • Can we use contextual data to better understand and model the economic impact of the epidemic and the corresponding reactive interventions to adjust response strategies?
  • Can contextual data help to more reliably predict the burden on the healthcare system and (re-)allocate resources?
  • Using region-specific data, can we more accurately model the effect of public health interventions such as social distancing or the closing of bars and restaurants?