|Inference for differential equation models with discretization uncertainty|
Ohio State University
Monday, December 8th, 2014
Exact inference for differential equation models requires the ability to evaluate states explicitly for given parameter values. However, model solutions are rarely available in closed form and many existing inferential tools therefore rely on time discretization and the resulting approximate likelihood. In the first part of this talk, I will show that ignoring discretization uncertainty often results in biased parameter estimates, even for apparently simple ODE and PDE models. The second part of this talk will introduce a new formalism for modelling and propagating this uncertainty through the Bayesian inferential framework, allowing exact inference and uncertainty quantification for discretized differential equation models.