Towards a Bayesian framework for model validation

Abstract

In this paper we discuss some concepts and a methodology of a Bayesian framework for model validation under uncertainty, which produces a probabilistic value for a models validity and may be used in the design of validation experiments. By using a stochastic metric as a measure of the distance between experiment and prediction, we update a validation distribution. We show this in practice using a simple numerical experiment and discuss the current shortcomings of the method. We finally discuss the role of information entropy in designing validation experiments.