Bayesian inference for discrete-time spatial epidemic processes скачать в хорошем качестве

Bayesian inference for discrete-time spatial epidemic processes

Speaker: Chris Jewell (CHICAS, Lancaster University) Date: Tuesday 23 March 2021, 4.00PM Location: Online Abstract The SARS-CoV-2 pandemic has highlighted the importance of spatial and temporal outbreak dynamics within highly heterogeneous human populations. In the UK, the epidemic has been characterised by regional foci that have occurred at different times in different places, and have often been prefaced by local “hotspots” of infection. Even with the vaccination campaign, detecting such spatial hotspots remains of critical importance for early intervention to stop small foci developing into larger outbreaks. The need for predictive (rather than merely descriptive) information on the outbreak requires the use of dynamical state-transition models, though these must be fitted in high spatial dimensions and in the presence of considerable data censoring. This talk will describe a Bayesian spatial SEIR model we are using to characterise the SARS-CoV-2 epidemic within the UK, and present details of a discrete-space MCMC-based method used for unbiased estimation of the disease transmission dynamics. In particular, it will focus on the problem of data augmentation for discrete-space models with counting-process constraints imposed by the epidemic model structure. Finally, the talk will outline some of the practical problems with embedding such an approach to inference and prediction within an automated pipeline given the requirements for high performance computing and data privacy.