Jornal article

International Journal of Approximate Reasoning

2008

3 (2008)

359-367

0888-613X

Prof. Radim Jiroušek, DrSc. (ÚTIA - Oddělení matematické teorie rozhodování)

Knowledge representation, Bayesian network, Operator of comp

The paper is a short review and comparison of two probabilistic models for uncertain knowledge representation: Bayesian networks and compositional models. These two approaches were chosen because they represent the same class of distributions and because they are typical representatives of the approaches using conditional (for Bayesian networks) and unconditional (for compositional models) distributions as basic building blocks for model construction. The comparison is made from the viewpoint of partial knowledge processing, in particular. Here we have in mind not only their capability to create global models from systems of pieces of local knowledge but most of all their efficiency to infer new pieces of local knowledge, different from those forming a generating (input) system.