Summary: A Tutorial on Learning With Bayesian Networks

We primarily summarize [4]. When we think that it is appropriate, we comment on additional facts and more recent developments. 1 Abstract and Introduction The advantages of graphical modelling include • easy handling of missing data • easy modelling of causal relationships • easy combination of prior information and data • easy to avoid overfitting 2 Bayesian Approach • Degree of belief • Rules of probability are a good tool to measure belief • Probability is subjective and has to be assessed for instance with a probability wheel. Problems of precision and accuracy can occur. • The paper considers as a running example multinomial sampling with Dirichlet priors. A Bayesian network (BN) is defined by • a network structure (DAG) • local probability distributions 1