Bayesian networks are a flexible class of models of data mining (but also of applied statistics). They can be used to capture the probabilistic dependency of variables and - contrary to pure prediction models as, e.g., decision trees - to predict varying and compound target variables. A bayesian net represents dependencies of variables by means of a graph and the exact quantities by probability tables.

The course presents an introduction to bayesian networks. Starting from modelling of (causal) influences and probabilities, we look at algorithms for exact and approximate inference (propagation of inference), analysis of bayesian networks, learning of parameters, and learning of structure.

Algorithms for inference and learning of bayesian networks rely heavily on graph algorithms, on common algorithms as topological sorting and checks for connection, as well as on more special methods as the enumeration of cliques etc. To keep the lecture as self-contained as possible, all required algorithms will be introduced during the course.