Exact and approximate inference in ProBT

We present a unifying framework for exact and approximate inference in Bayesian networks. This framework is used in "ProBT", a general purpose inference engine for probabilistic reasoning and incremental model construction. This paper is not intended to present ProB T but to describe its underlying algorithms mainly the "Successive Restrictions Algorithm " (SRA) for exact inference, and the "Monte Carlo Simultaneous Estimation and Maximization" (MCSEM) algorithm for approximate inference problems. The main idea of ProBT is to use "probability expressions " that can be "exact" or "approximate" as basic bricks to build more complex models incrementally.