Winners of the XPRIZE Pandemic Response Challenge

TL;DR: ValenciaIA4COVID won first place in the global $500K XPRIZE Pandemic Response Challenge. Its predictor estimated COVID-19 cases under different restrictions, while its prescriptor suggested plans that kept both infections and the cost of those restrictions as low as possible.

In March 2021, the ValenciaIA4COVID team won first place in the XPRIZE Pandemic Response Challenge, a four-month global competition sponsored by Cognizant. More than 100 teams entered the challenge, and 48 finalists from 17 countries were selected. ValenciaIA4COVID and Slovenia's JSI vs COVID shared the $500,000 prize purse, with the judges awarding first-place honors to the Valencian team for its exemplary submission.

The team brought together researchers from universities and research centers across the Valencian Region. It was coordinated by Nuria Oliver, Scientific Director of ELLIS Alicante and then Commissioner for the Valencian AI Strategy and Data Science against COVID-19, and J. Alberto Conejero of the Universitat Politècnica de València.

The challenge

Teams had to solve two problems:

  1. Predict the evolution of COVID-19 cases under different combinations of non-pharmaceutical interventions, such as limits on gatherings, school closures, workplace measures, or travel restrictions.
  2. Prescribe combinations of measures that kept both infections and the social and economic cost of the measures as low as possible.

The models went through quantitative tests and an independent review by specialists in epidemiology, AI, and pandemic response. The judges also looked at whether the systems were practical, usable, fair, open, and innovative.

The winning project

ValenciaIA4COVID trained a deep-learning predictor on case numbers and the measures adopted in different countries and regions. Given a proposed set of measures, it estimated how the number of cases would change.

Diagram of the ValenciaIA4COVID predictor, which combines case data and non-pharmaceutical interventions to forecast cases
The predictor estimates future case numbers from epidemiological data and proposed interventions.

The prescriptor used those forecasts to search for intervention plans. It did not return one answer. Instead, it produced a range of options: some prioritized fewer infections, while others used less restrictive measures.

Diagram of the ValenciaIA4COVID prescriptor, which compares intervention plans by expected cases and intervention cost
The prescriptor identifies intervention plans with different balances between expected cases and intervention costs.

The work did not start with the competition. The team had been collaborating with the Generalitat Valenciana since March 2020 on mobility analysis, computational epidemiology, forecasting, and citizen science. XPRIZE tested the models with real data, and the collaboration with the Generalitat provided a second real-world setting in which to validate them.

After the challenge

The team described the predictor, prescriptor, and validation results in Open Data Science to Fight COVID-19: Winning the 500K XPRIZE Pandemic Response Challenge. The paper received the Best Applied Data Science Paper Award at ECML-PKDD 2021. A second version appeared in the IJCAI 2022 proceedings.

Team members

The ValenciaIA4COVID team included:

  • Kristina Polotskaya
  • Òscar Garibo
  • Nuria Oliver
  • Manuel G. Portolés
  • Miguel Angel García
  • Victor de Elena
  • J. Alberto Conejero
  • Miguel Rebollo
  • Francisco Escolano
  • Elisa Espín
  • Eloy Piñol Jimenez
  • Aurora Mula Leal
  • Adolfo Lopez
  • Miguel Angel Lozano

The team included researchers affiliated with ELLIS Alicante, Universitat Jaume I, Universidad de Alicante, Universidad Miguel Hernández, Universitat Politècnica de València, and Universidad Cardenal Herrera CEU.