Stats4Grads: Emulating Multiple Computer Models
9 February 2011 14:15 in CM219
Although complex computer models can be useful and informative in our understanding of complicated systems, in practical terms their use is limited by a lack of knowledge and understanding and by their computational intensity. In this talk we attempt to address both of these problems using Bayesian Emulators. Much work has been done, in Durham and elsewhere, on building emulators to approximate complex models, and they enable us to accurately approximate the models in a much more efficient and tractable way. This, to an extent, deals with the computational aspect of the problem. However, the emulators can only be as informative for the system as the model itself, and where the model lacks or misrepresents information the emulator will also. One way to address this is to combine several models of the same system, all of which will be accurate and informative for different aspects of the system. Here, we make first steps toward emulating several complex models at once.
See the Stats4Grads page for more details about this series.