NeurIPS 2020 Highlights AI Research Pushing to Solve the COVID-19 Crisis
NeurIPS 2020 features a dedicated track of accepted papers where machine learning models tackle the global pandemic. Researchers present innovative data-driven approaches to predict outcomes and assess lockdown policies.
NeurIPS 2020 features a special selection of research papers dedicated to tackling the COVID-19 pandemic. Out of forty submitted papers on this topic, the conference accepts one for an oral presentation, four for spotlight presentations, and four for poster presentations. This focused effort shows how the artificial intelligence community rapidly mobilizes to address the global health crisis.
One standout paper from researchers at the University of Cambridge and UCLA develops a Bayesian model to assess global lockdown policies. The system treats each country as a distinct data point and uses variations in government policies to learn country-specific effects. This approach allows the model to predict COVID-19 fatalities under different containment scenarios.
The predictive model relies on a two-layer Gaussian process prior built on a compartmental SEIR framework. The lower layer uses country-and-policy-specific parameters to capture fatality curves under counterfactual policies, while the upper layer shares data across all countries to learn parameters based on national features and policy indicators. This architecture combines solid epidemiological mechanics with advanced machine learning to help governments make informed decisions about lifting lockdowns.