Method(olog)ische Grundlagen der Statistik und ihre Anwendungen
print


Navigationspfad


Inhaltsbereich

Oberseminar

Wir treffen uns unregelmäßig um in einem informellen Rahmen über Themen zu diskutieren, welche die Fundierung der Statistik tangieren, wie beispielsweise statistisches Schließen, das Treffen von Entscheidungen unter Unsicherheit oder das Gebiet der Imprecise Probability.

Interessierte sind herzlich eingeladen!

Präsentationen in einem formelleren Rahmen und eingeladene Vorträge werden auf diesen Seiten gesondert angekündigt.

Kontaktieren Sie Paul Fink für weitere Informationen.

Vorträge

DatumThema
27. Juni 2016

Jürgen Landes (MCMP): Objective Bayesian Nets from Consistent Datasets

This paper addresses a data-integration problem: given several mutually consistent datasets each of which measures a subset of the variables of interest, how can one construct a probabilistic model that fits the data and gives reasonable answers to questions which are underdetermined by the data? Here we show how to obtain a Bayesian network model which approximates the unique probability function that agrees with the probability distributions induced by the datasets and otherwise has maximum entropy. We give a general algorithm which is significantly more efficient than the standard brute-force approach for determining the maximum entropy probability function.

We report on a Matlab implementation of this general algorithm as evidence for the computational superiority of the general algorithm over the brute force approach. Furthermore, we show that in a wide range of cases such a Bayesian net representation can be obtained by solving an algebraic problem rather than a computationally much harder optimisation problem. For other cases we show how to obtain the Bayesian net by solving a comparatively simple optimisation problem.