Modeling First-Order Bayesian Networks (FOBN)

Bayesian networks provide an elegant formalism to perform inferences under uncertainty. Their shortcoming of being propositional in nature, however, restricts their expressive power and restrains their use in domains where number of instances may vary from situation to situation. First-order Logic (FOL), on the other hand, enjoys that power of expressiveness but is deterministic in nature. Integration of Bayesian networks and first-order logic provides powerful mechanism to capture and process domains that are truly dynamic and non-deterministic. The paper explores and compares three different probabilistic languages, namely Bayesian Logic Program (BLP), Bayesian Logic (BLOG) and Multi-Entity Bayesian Network (MEBN) that provide support to develop First Order Bayesian Networks (FOBN). The study identifies key characteristics that are prevalent in all three languages and compares their relative strengths and weaknesses.