Michael Whitehouse (Imperial)

Speaker: Michael Whitehouse (Imperial)

Title: Scalable calibration for partially observed individual-based epidemic models through categorical approximations

Abstract: Traditional compartmental models successfully capture population-level dynamics but fail to characterize individual-level risk. The computational cost of exact likelihood evaluation for partially observed individual-based models, however, grows exponentially with the population size, necessitating approximate inference. Existing sampling-based methods usually require multiple simulations of every individual in the population, are heavily reliant on bespoke proposal distributions or summary statistics, and scale poorly with population size. To address these issues, we propose a deterministic approach to approximating the likelihood using categorical distributions. The resulting algorithm has a computational cost as low as linear in population size. The approximate likelihood is amenable to automatic differentiation, so parameters can be estimated by maximization or posterior sampling using standard software libraries such as Stan or TensorFlow with little user effort. We prove consistency of the maximum approximate likelihood estimator. We empirically test our approach on models with various flavours of heterogeneity: different sets of disease states, individual-specific transition rates, heterogeneous spatial mixing, under-reporting and misreporting. We demonstrate ground truth recovery and comparable marginal log-likelihood values at substantially reduced cost compared to competitor methods. Finally, we demonstrate the scalability and effectiveness of our approach with a real-world application using data from the 2001 UK Foot-and-Mouth outbreak.