LEARNING AND REPRESENTING CAUSAL RELATIONSHIPS WITH BAYESIAN NETWORKS

摘要

Bayesian networks axe graphical tools for modeling probabilistic relationships between variables of a system. When those relationships are causal relationships the Bayesian network is also called a causal Bayesian network. This paper provides an overview of Bayesian networks and causal Bayesian networks, and also a brief look at some of the algorithms for performing exact and approximate inference, and for learning Bayesian networks from data, which involves learning the structure or causal structure of the network and then learning the probability distributions of the variables. A Java-based implementation of some of the algorithms and the experimental evaluation results are also discussed.