Confidence Analysis for Nuclear Arms Control: SMT Abstractions of Bayesian Belief Networks

摘要: How to reduce, in principle, arms a verifiable manner that is trusted by two or more parties hard but important problem. Nations and organisations wish engage such control verification activities need be able design procedures mechanisms capture their trust assumptions let them compute pertinent degrees of belief. Crucially, they also will methods for reliably assessing confidence computed belief situations with little no contextual data. We model an scenario what we call constrained Bayesian Belief Networks cBBN. A cBBN represents set symbolically expressing uncertainty about probabilities scenario-specific constraints are not represented BBN. show this abstraction BBNs can mitigate well against the lack prior Specifically, describe how cBBNs have faithful representations within Satisfiability Modulo Theory SMT solver, these open up new ways automatically may cBBNs. Furthermore, perform symbolic sensitivity analyses cBBNs, global optima under-specified particular interest decision making. solving enables us assess relative same scenario, where models share some information express aspects at different levels abstraction.