The multiinformation function as a tool for measuring stochastic dependence

The topic of the talk is the concept of mutliinformation, which is one of information-theoretical characteristics of the degree of stochastic dependence among (finitely many) random variables. In the first part of the talk, a result about the non-existence of an axiomatic characterization of conditional indpendence models is recalled as an example of the theoretical use of the multiinformation function. In the second part, the multiinformation is related to conditional mutual information, which complies with several reasonable requirements on a measure of the degree of stochastic conditional dependence. The third part of the talk responds to an interesting question whether it is possible to decompose multiinformation into level-specific measures of dependence among variables.