Abstract
philosophies and methods that can be used to select models. In this work, we discuss the application of several of these methods to a set of models which are underdetermined by the available data. Model selection is typically framed as a balance between accuracy and complexity, however, this scenario poses a dierent set of challeneges because all the models can be considered accurate. We apply the most common Frequentist and Bayesian model selection methods to a nuclear reactor model using the R7 code. In the process the DAKOTA framework is used for both classical and Bayesian calibration and building Gaussian Process surrogate models. We nd that Frequentist approaches are of little use for selecting between underdetermined models and the Bayesian approach does not fare much better. We conclude with some insight into the use of Frequentist or Bayesian methods for this scenario and how the results can be improved by using better quality data. In this case, the Bayesian approach provides some clear advantages over the Frequentist methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have