Causality enabled compositional modelling of Bayesian networks

Probabilistic abduction extends conventional symbolic abductive reasoning with Bayesian inference methods. This allows for the uncertainty underlying implications to be expressed with probabilities as well as assumptions, thus complementing the symbolic approach in situations where the use of a complete list of assumptions underlying inferences is not practical. However, probabilistic abduction has been of little use in first principle-based applications, such as abductive diagnosis, largely because no methods are available to automate the construction of probabilistic models, such as Bayesian networks (BNs). This paper addresses this issue by proposing a compositional modelling method for BNs.