Breadcrumb
Inference for complex physical systems
Supervisor: Jonty Rougier
Theme: Bayesian Modelling & Analysis
Complex physical systems are now routinely simulated on computers, ranging all the way from protein folding at the molecular level to the galaxy formation at the cosmic level. Where statistics is involved these types of study are known as 'Computer Experiments'.
There are two main statistical challenges if we want to use the evaluation(s) of a computer model to make inferences about the underlying system. The first is how we account for structural errors in the model; the second, how best we can use the limited resources at our disposal, given that many models are expensive to evaluate.
I have a particular interest in climate prediction. This is a good test-bed for statistical methodology and practice, because it combines both challenges: even the largest climate models are still poor at representing transient and regional behaviour, and they are also among the largest and most expensive models every built.