Parameter estimation in expensive computational models
Umberto Noe (University of Glasgow)
Friday 2nd June, 2017 15:00-16:00 To be announced soon
In this talk we discuss and show methods to perform inference in expensive computational models, for example involving the numerical solution of a system of PDEs. These models are referred to as "black-box" functions and typically are not suitable for MCMC or standard likelihood based inference due to the high computational resources needed for a single evaluation. We will compare and contrast the concepts of Emulation and Bayesian Optimization, as well as discuss how to extend their application to scenarios where the computational model has unknown regions in the domain where the model assumptions do not hold and returns a failed simulation (hidden constraints).