All Times EDT
Virtual
A Hamiltonian Monte-Carlo Sampling Scheme for Bayesian Compartmental Epidemic Models (309904)
Grant Brown, University of Iowa*Helin Hernandez, University of Iowa
Jacob Oleson, University of Iowa
Keywords: Hamiltonian Monte Carlo, stochastic compartmental models, infectious disease modeling
Stochastic compartmental models are useful for understanding the complexity of infectious diseases but are computationally burdensome due to high-dimensional parameter spaces. Most commonly, stochastic epidemic models are implemented via a Bayesian paradigm, where parameters of interest can be estimated using MCMC. However, standard MCMC algorithms can suffer from high autocorrelation and inefficient exploration of the parameter space when the target distribution is high-dimensional, as is often the case in infectious disease modeling. This study investigated the performance of the Hamiltonian Monte Carlo (HMC) proposals for parameters in SEIR models for infectious diseases. Although the HMC algorithm is more expensive per iteration compared to the Metropolis-Hastings algorithm, it can traverse the parameter space more efficiently. Sampling schemes were compared in terms of convergence and effective samples generated per minute. We show that HMC can produce better convergence and increased efficiency for key epidemic parameters.