Syllabus
The concept of conditional independence (CI). Basic formal properties of CI, the concept of a semi-graphoid and (formal) CI structure. Basic method of construction of measures inducing CI structures. Information-theoretical tools for CI structure study. Graphical methods for CI structure description: undirected graphs (= Markov networks), acyclic directed graphs (= Bayesian networks). The method of local computation.
Possible additional topics: The (non-existence of a) finite axiomatic characterization of CI structures. Learning graphical models from data. Chain graphs.
Annotation
The lecture is conceived as an introduction to the above mentioned topic and it leads to the methods of
(mathematical) description of probabilistic conditional independence (CI) structures by means of tools of discrete mathematics, in particular by means of graphs whose nodes correspond to random variables. Because CI structures occur both in modern statistics and in artificial inteligence (so-called probabilistic expert systems) the lecture is suitable both for students of probability and statistics and for the students of informatics.