Speaker: Jason Xu, Assistant Professor, Department of Statistical Science, Duke University

Title: Likelihood-based inference for stochastic epidemics via data-Augmented MCMC 

Abstract: We propose novel data-augmented Markov Chain Monte Carlo strategies to enable exact Bayesian inference under the stochastic susceptible-infected-removed model and its variants. In the incidence data setting, where we are given only discretely observed counts of infection, significant challenges to inference arise due only a partially informative glimpse of the underlying continuous-time process. To account for the missing data while targeting the exact posterior of model parameters, we make use of latent variables that are jointly proposed from  surrogates related to branching processes, carefully designed to closely resemble the SIR model. This allows us to efficiently generate epidemics consistent with the observed data, and extends to non-Markovian settings as well as tasks such as simultaneous change-point detection under time-varying transmission. Our Markov chain Monte Carlo algorithm is shown to be uniformly ergodic, and we find that it mixes significantly faster than existing single-site samplers on several real and simulated data applications.