Konferenční příspěvek

Kratochvíl V., Jiroušek R.

Information Processing and Management of Uncertainty in Knowledge-based Systems.

3

Éditions EDK

Paris

2006

2300-2307

2-84254-112-X

Ing. Václav Kratochvíl (ÚTIA - Oddělení matematické teorie rozhodování)

Compositional model, Multidimensional distribution

Kratochvíl, Jiroušek: IPMU 2006, Marginalization Algorithm for Compositional Models [file]

The paper deals with a problem of marginalization of multidimensional probability distributions represented by compositional models, more precisely by perfect sequence models. It appears that from the computational point of view, the solution is more efficient than any known marginalization process for Bayesian networks. This is because the process, which is in the paper described in a form of an algorithm, takes advantage of the fact that perfect sequence models have some information explicitly encoded, which can be got from Bayesian networks only by application of rather computationally expensive procedures.