The learning of high dimensional Gaussian probability networks is an important field of artificial intelligence research. These networks are optimal for representing the relations between variables because they use the local structure of the problem.
These networks are widely used for reconstructing gene regulatory networks which are especially important part of nowadays’ research in biomedical sciences. Measured high dimensional gene expressional data is getting more available due to public databases (eg. Gene Expression Omnibus) which makes the usage of such networks even more possible. The regulation of genes is a very complex process in which not just the regulated genes, but also the environment plays an important role.
We know about existing efficient algorithms for learning the parameters of purely Gaussian probability networks, but learning networks with mixed, discrete and continuous variables is still an active research field. A possible way of solution is for example is using neural networks in the nodes for calculating the effect of the parent nodes and the environmental variables.