Seminar in applied mathematics and statistics

SPEAKER: Peter Harremoës (Copenhagen Business College)

TITLE: Causation via Lattice Theory

ABSTRACT: For the last 30 years Bayesian networks and similar graphical models have been popular models  for modelling cause, effect, influence and related concepts. These graphical models are often presented as models for conditional independence. In this talk the focus wil be on functional dependencies instead of conditional independence. Functional dependences can be described by lattices theory and in this talk we shall see how for instance causal loops translate into lattice language. This approach has some highly non-trivial applications in information theory (and group theory!) and it also suggests that the graphical models is a too narrow class of statistical models for some problems. It may also serve as an alternative foundation for causal inference. The talk will not require any prior knowledge of lattice theory.

Tea and chocolate will be served in room 04.4.19 after the seminar.

------

Upcoming seminars:

April 6, 13:15: Kayvan Sadeghi (Cambridge)