Probabilistic reasoning as quadratic unconstrained binary optimization

Probabilistic reasoning is an important tool for using uncertainty in AI, especially for automated reasoning. Partial probability assessments are a way of expressing partial probabilistic knowledge on a set of events. These assessments contain only the information about "interesting"events (hence it can be easily assessed by an expert). On the other hand, partial assessments can cause consistency problems. In this paper we show how to formulate the main tasks of probabilistic reasoning on partial probability assessments, namely check of coherence, correction, and inference, as QUBO problems. This transformation allows to solve these problems with a quantum or a digital annealer and thus providing new computational methods to perform these tasks.