Two Conditions for Modelling Large Bayesian Networks

Seong-Ho Kim and Sung-Ho Kim (Korea)


Consistency of distribution; D-split; EM; Family condition; Hyper-EM condition; Join tree of submodels; Node remov ability; t-split


Consider building a large Bayesian network of categorical variables by model-splitting as proposed in Kim [7]. We propose two conditions so that the hyper-EM of Kim [7] be applied for modelling Bayesian networks of any size in a computationally efficient way. Although we considered Bayesian networks of categorical variables only, the result can be extended to any type of variables.

Important Links:

Go Back