Inference in Network-Based Epidemiological Simulations with Probabilistic Programming

Accurate epidemiological models require parameter estimates that account for mobility patterns and social network structure. This work applies probabilistic programming to infer parameters in agent-based models. We represent mobility networks as degree-corrected stochastic block models and estimate their parameters from cell-phone co-location data. We use these networks in probabilistic programs to simulate the evolution of an epidemic, and condition on reported cases to infer disease transmission parameters. Our experiments demonstrate that the resulting models improve the accuracy-of-fit in multiple geographies relative to baselines that do not model network topology.

Article Link

https://aiforpublichealth.github.io/papers/ICLR-AI4PH_paper_36.pdf

Bibtex

@inproceedings{covid-param-inference,
  title={Inference in Network-based Epidemiological Simulations with Probabilistic Programming},
  author={
  Smedemark-Margulies, Niklas and
  Walters, Robin and
  Zimmermann, Heiko and
  Laird, Lucas and
  Kaushik, Neela and
  Caceres, Rajmonda and 
  van de Meent, Jan-Willem},
  booktitle={AI for Public Health Workshop, ICLR},
  year={2021},
  url={https://aiforpublichealth.github.io/papers/ICLR-AI4PH_paper_36.pdf}
}