An in-depth case study: modelling an information barrier with Bayesian Belief Networks

Abstract

We present in detail a quantitative Bayesian Belief Network (BBN) model of the use of an information barrier system during a nuclear arms control inspection, and an analysis of this model using the capabilities of a Satis ability Modulo Theory (SMT) solver. Arms control veri cation processes do not in practice allow the parties involved to gather complete information about each other, and therefore any model we use must be able to cope with the limited information, subjective assessment and uncertainty in this domain. We have previously extended BBNs to allow this kind of uncertainty in parameter values (such as probabilities) to be re ected; these constrained BBNs (cBBNs) o er the potential for more robust modelling, which in that study we demonstrated with a simple information barrier model. We now present a much more detailed model of a similar veri cation process, based on the technical capabilities and deployment concept of the UK-Norway Initiative (UKNI) Information Barrier system, demonstrating the scalability of our previously-presented approach. We discuss facets of the model itself in detail, before analysing pertinent questions of interest to give examples of the power of this approach.