Optimal design of experiments optimizes experimental schemes to minimize the uncertainty of model predictions. An experimental scheme refers to the choice of sampling locations, data types and system excitation to trigger and observe system responses.From the information-theoretic point of view, a sampling design is as informative on the prediction goal as the prediction goal is on the sampled data. This project will revert the flow of information between data and prediction goal. The new approach will lead to a dramatic reduction of computational costs, and therefore allow to use much more accurate conditioning methods than usual.With smaller costs and more accurate schemes, larger, more complex and more non-linear systems and problems can be tackled. Exemplary applications in this project will be the investigation of contaminated sites and model selection problems.
08/2009 - 02/2014