A Total Entropy Criterion for the Dual Problem of Model Discrimination and Parameter Estimation

Abstract

Summary A total entropy criterion is developed for the sequential design of experiments. The criterion is applicable to the dual problem of model discrimination and parameter estimation. The total entropy measures both the uncertainty about which mathematical model is correct and the uncertainty about the parameter vector for each model. The criterion is shown to lead to the choice of experiment for which the outcome is most uncertain, relative to the uncertainty due to experimental error.