In our upcoming Bayesian Working group meeting  tomorrow, Monday (04/21), 1:10 - 2 pm in Snedecor 2113, Sebastian McCrimmon will be presenting.

Title: Bayesian Emulation of High Dimensional Multifidelity Computer Models

Abstract:

Researchers across many scientific fields often study complicated natural processes using computer models that produce a large number of outputs. Many models can be implemented over a range of fidelity levels. Greater precision typically increases the computational burden by several orders of magnitude, necessitating the development of an emulator which leverages information from low-fidelity outputs to provide fast and accurate predictions of high-fidelity simulations. In this talk, we discuss a Bayesian hierarchical modeling approach for emulating multifidelity computer models that produce a large number of outputs over a spatial domain. We consider a multivariate autoregressive cokriging emulator with a separable covariance structure. The modeling approach is illustrated using a deterministic model of pollutant concentration after a chemical spill taken from the literature.

Look forward to seeing many of you there.