A0332
Title: Markov bases from discrete to continuous frameworks
Authors: Fabio Rapallo - University of Genova (Italy) [presenting]
Abstract: The notions of Markov moves and Markov bases from Algebraic Statistics are traditionally defined in a discrete framework to define a connected Markov chain on the set of contingency tables under linear constraints, thus working with empirical distributions of counts. In the context of rater agreement analysis, Markov bases can be used to find the maximum value of kappa-type statistics through a simulated annealing algorithm. The same objects and the same algorithm can also be applied in a continuous setting, i.e., working with probability distributions, to find the Kantorovich distance between two distributions on a finite sample space. Here we highlight the analogies and the differences between the two approaches.
