We consider the problem of generating a joint distribution for a pair of Bayesian networks that preserves the multivariate marginal distribution of each network and satisfies prescribed correlation between pairs of nodes taken from both networks. We derive the maximum entropy distribution for any pair of multivariate random vectors and prescribed correlations and demonstrate numerical results for an example integration of Bayesian networks.
Revised: March 26, 2012 |
Published: August 30, 2011
Citation
Jarman K.D., and P.D. Whitney. 2011.Integrating Correlated Bayesian Networks Using Maximum Entropy.Applied Mathematical Sciences 5, no. 48:2361-2371.PNNL-SA-75301.