Thursday 30 Jan 2020: Efficient Bayesian model choice for partially observed (infectious disease) processes
Dr TJ McKinley -
Harrison 101 14:30-16:00
Mathematical and statistical models can provide key insights into the mechanisms that underlie complex processes, such as the spread and persistence of infectious diseases. However, their utility is linked to the ability to adequately calibrate these models to observed data. Performing robust inference for these systems is challenging, since the models often exhibit complex non-linear dynamics, and available data are almost always incomplete, thus rendering the likelihood intractable.
Simulation-based inference techniques, such as pseudo-marginal MCMC, have shown great promise in this area. These utilise simulation models to numerically integrate over the missing information, and are easier to code and generalise than alternative methods such as data-augmentation and/or reversible-jump MCMC. In this talk we highlight the utility of these approaches for inference, and introduce ways in which they can also be used for efficient Bayesian model comparison. We also discuss key challenges that remain. We illustrate these approaches on data from an experimental transmission study of highly pathogenic avian influenza in genetically modified chickens.