Statistics Seminar :: Bayesian Validation of Computer Models
Computer models are mathematic representations of real systems developed for understanding and investigating the systems. They are particularly useful when physical experiments of the systems are either cost prohibitive or time prohibitive. It is very important to validate computer models by comparing computer outputs and physical observations before the computer models are used. This paper proposes a Bayesian approach for computer model validation. The proposed approach overcomes several difficulties of a frequentist approach proposed in Oberkampf and Barone (2004). Kennedy and O'Hagan (2001) propose a similar Bayesian approach for the calibration problem in computer model validation. The major difference between the two approaches is that Kennedy and O'Hagan(2001) focus on deriving directly the posterior of the true output. Our approach focuses on first deriving the two posteriors of the computer model and the model bias (difference between computer and true outputs), and then derive the posterior of the true output as the sum of the two posteriors. As a result, our approach provides a clear decomposition of the expected prediction error of the true output. This decomposition explains why and how combining both computer output and physical experiments can provide more accurate prediction of the true output than using only computer outputs or only physical experiments. Three examples are used to illustrate the proposed approach and to compare with the approach in Kennedy and O'Hagan.
- Workflow Status: Published
- Created By: Barbara Christopher
- Created: 10/08/2010
- Modified By: Fletcher Moore
- Modified: 10/07/2016