Bayesian Rating Systems

Пятница, 25 ноября, к. 106. Начало в 18:00.

Докладчик: С.И. Николенко (ПОМИ РАН).

Тема: Bayesian Rating Systems.

Abstract

Bayesian rating systems are probabilistic models that infer a linear ordering on a set of entities (i.e., their ratings) based on the results of numerous noisy comparisons of small subsets (i.e., individual matches/tournaments). In the talk, I will introduce Bayesian rating systems and present the construction of TrueSkill (Graepel, Minka, Herbrich, 2007) and its recently developed improved version (Nikolenko, Sirotkin, 2011).