TB MAC seminar series: Calibrating complex models with emulation & history matching (Andy Iskauskas)

Seminar summary:
Complex models of disease transmission and dynamics are paramount in investigating the impact of interventions or making predictions of future outcomes. However, their complexity can be a barrier to an exhaustive exploration of parameter space, whether as a result of stochasticity, high-dimensional input space, or computational expense of simulator runs. To surmount this problem one may use the twin methodologies of emulation (a means by which a potentially slow simulator can be represented by a fast statistical surrogate) coupled with history matching (an exploration method that finds acceptable matches to observational data in the presence of uncertainty). Even where these techniques are well-defined, it may be difficult to implement such methods into a standard calibration workflow: I will demonstrate how the R package hmer enabled their application to a complex deterministic TB model across multiple countries, and discuss built-in extensions for stochastic simulators.

Presenter bio:
Andy Iskauskas is an Assistant Professor in Statistics and Willmore Fellow at Durham University, with particular focus on Bayesian calibration. He has worked with epidemiologists across the world modelling a number of different diseases, as well as non-epidemiological applications in physics, biology, and industry, and developed and maintains the history matching and emulation package hmer.

Скачать видео