Statistics Seminars: Likelihood inference for computer models of complex physical systems
8 March 2010 14:00 in CM221
Statistical analysis of computer code implementation of deterministic models of complex physical systems, such as those for climate, hydrocarbon reservoirs and galaxy formation, is a growing field. There is a sizeable group in the department working in this area, some of them employed on the UK Research Councils supported multi-centre four-year project `Managing Uncertainty for Computer Models'.
The talk will cover emulation of multivariate computer codes and likelihood inference for (i) correlation lengths of Gaussian processes, (ii) Box-Cox transformations, (iii) the computer code input which 'best' matches observations on the physical system, and (iv) a parameterised covariance structure of the discrepancy between the computer code and the physical system.
The methods will be illustrated with data from one or more applications. Robustness of likelihood inference to usual Gaussian assumptions for will be considered.
Contact firstname.lastname@example.org for more information