Statistics Seminars: Estimating and emulating internal discrepancy for computer simulators
25 January 2016 14:00 in CM221
Computer simulators are an important and widely-used tool in understanding complicated systems. They are, however, only models for reality, and so even at well-calibrated parameter settings the simulator will not make perfect predictions. The discrepancy between the simulator output and the real system must be accounted for when using simulators. Some aspects of this discrepancy can be explored by performing perturbation experiments on the simulator. I will outline the nature of these experiments and show how the Bayes linear framework can be used to make inferences from them about simulator discrepancy for a particular choice of parameters. Then, I will show how perturbation experiments at several different parameter settings can be combined to make inferences about the simulator discrepancy for all parameter settings, using the popular emulation method.
Contact firstname.lastname@example.org for more information