Relationship between properties characterizing independence equivalence in Bayesian networks and compositional models

0353102 - ÚTIA 2011 JP eng C - Conference Paper (international conference)
Kratochvíl, Václav
Relationship between properties characterizing independence equivalence in Bayesian networks and compositional models.
Proceedings of the 13th Czech-Japan Seminar on Data Analysis and Decision Making in Service Science. Otaru: University Hall, 2010 - (Itoh, T.; Suzuki, K.), s. 49-54
[Czech-Japan Seminar on Data Analysis and Decision Making under Uncertainty /13./. Otaru (JP), 03.11.2010-05.11.2010]
R&D Projects: GA MŠMT 1M0572; GA ČR GA201/09/1891
Grant - others:GA ČR(XE) ICC/08/E010 Eurocores LogICCC; GA MŠk(CZ) 2C06019
Institutional research plan: CEZ:AV0Z10750506
Keywords : compositional model * bayesian network * equivalence problem
Subject RIV: IN - Informatics, Computer Science
http://library.utia.cas.cz/separaty/2010/MTR/kratochvil-relationship between properties characterizing independence equivalence in bayesian networks and compositional models.pdf

We present a simple characterizations of equivalent compositional model structures based on (A) invariant properties and (B) local transformations. It has been shown that one can transform any compositional model into Bayesian network representing the same joint probability distribution and vice versa. Moreover, every assertion of independence induced by a Bayesian network structure is also induced by the structure of the respective compositional model that is created from the Bayesian network and vice versa. That is why wa can simply compare characterization of equivalent compositional model structures together with the known characteristics of equivalent Bayesian network structures. We show which (A) invariant properties and (B)local transformations correspond each other. In opposite case we show what is its (invariant property, transformation) meaning in the other model structure.
Permanent Link: http://hdl.handle.net/11104/0192441