Bayesian Networks

A Bayesian network is a graphical representation of a multivariate probability distribution on a set [Page 64]of discrete random variables. Representational efficiency is achieved by explicit separation of information about conditional independence relations between the variables (coded in the network structure) and information about the probabilities involved (coded as a set of numeric parameters or functions). The network structure is expressed as a directed acyclic graph (DAG) that makes the representation amenable to an intuitively appealing, causal interpretation. Algorithms exist for learning both network structure and parameters from data. Furthermore, Bayesian networks allow for computing any marginal or conditional probability regarding the variables involved, thus offering a powerful framework for reasoning with uncertainty. Bayesian networks are also called belief networks and causal probabilistic networks.