Developing a Validation Methodology for Expert-Informed Bayesian Network Models Supporting Nuclear Nonproliferation Analysis

Conference · Tue May 13 00:00:00 EDT 2014

OSTI ID:1236342

White, Amanda M.; Gastelum, Zoe N.; Whitney, Paul D.

Under the auspices of Pacific Northwest National Laboratory’s Signature Discovery Initiative (SDI), the research team developed a series of Bayesian Network models to assess multi-source signatures of nuclear programs. A Bayesian network is a mathematical model that can be used to marshal evidence to assess competing hypotheses. The purpose of the models was to allow non-expert analysts to benefit from the use of expert-informed mathematical models to assess nuclear programs, because such assessments require significant technical expertise ranging from the nuclear fuel cycle, construction and engineering, imagery analysis, and so forth. One such model developed under this research was aimed at assessing the consistency of open-source information about a nuclear facility with the facility’s declared use. The model incorporates factors such as location, security and safety features among others identified by subject matter experts as crucial to their assessments. The model includes key features, observables and their relationships. The model also provides documentation, which serves as training materials for the non-experts.

• Research Organization: Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
• Sponsoring Organization: USDOE
• DOE Contract Number: AC05-76RL01830
• OSTI ID: 1236342
• Report Number(s): PNNL-SA-102659
• Resource Relation: Conference: Proceedings of the Information Analysis Technologies, Techniques and Methods for Safeguards, Nonproliferation and Arms Control Verification Workshop, May 12-14, 2014, Portland, Oregon, 229-236
• Country of Publication: United States
• Language: English