This title appears in the Scientific Report :
2018
Approximate Bayesian Computation Applied to Nuclear Safeguards Metrology
Approximate Bayesian Computation Applied to Nuclear Safeguards Metrology
Approximate Bayesian Computation (ABC) is an inference option if a likelihood for measurement data is not available, but a forward model is available that outputs predicted observables, such as gamma counts, for any set of specified input parameters, such as item mass. This paper reviews ABC...
Saved in:
Personal Name(s): | Burr, Tom Lee (Corresponding author) |
---|---|
Krieger, Thomas / Norman, Claude | |
Contributing Institute: |
Nukleare Entsorgung; IEK-6 |
Published in: | ESARDA bulletin, 57 (2018) S. 50-59 |
Imprint: |
Ispra
ESARDA
2018
|
Document Type: |
Journal Article |
Research Program: |
Joint Programme on the Technical Development and Further Improvement of IAEA Safeguards between the Government of the Federal Republic of Germany and the International Atomic Energy Agency Nuclear Waste Management |
Publikationsportal JuSER |
Approximate Bayesian Computation (ABC) is an inference option if a likelihood for measurement data is not available, but a forward model is available that outputs predicted observables, such as gamma counts, for any set of specified input parameters, such as item mass. This paper reviews ABC and illustrates how ABC can be applied in safeguards metrology. A key aspect of metrology is uncertainty quantification (UQ), approached from physical first principles (“bottom-up”) or approached empirically by comparing measurements from different methods and/or laboratories (“top-down”). Although ABC is not yet commonly used in metrology, an example using enrichment measurements is used to illustrate potential advantages in ABC compared to current bottom-up approaches. Using the same example, ABC is also shown to be useful in top-down UQ. And, the example shows good agreement between bottom-up and top-down measurement error relative standard deviation (RSD) estimates, while also allowing for the effects of item-specific biases. As a diagnostic, in applications of ABC, the actual coverages of probability intervals are compared to the true coverages. For example, if an ABC-based interval for the true measurement RSD is constructed to contain approximately 95% of the true values, then one can check whether the actual coverage is close to 95%. It is shown that one advantage of ABC compared to other Bayesian approaches is its apparent robustness to miss-specifying the model while maintaining good agreement between the nominal and the actual coverage. |