Universidad Complutense de Madrid :: NILS mobility project

"The objective of the stay is to continue and strengthen our collaboration on inference in hybrid Bayesian networks using Mixtures of Truncated Exponentials (MTEs):

Bayesian networks constitute one of the most successful frameworks for automatic reasoning under uncertainty. However, a difficulty with the frame work is its strict requirements regarding how to structure the domain under study. In particular, so-called hybrid domains (domains containing both discrete and continuous variables) pose a challenge to state-of-the-art inference algorithms. A solution pursued earlier (through, e.g., the ELVIRA Project and the ProGraMo Project) is to approximate the required densities by a family of functions known as the Mixture of Truncated Exponentials (MTEs). MTEs can approximate any distribution arbitrarily well, and is also important as it is closed under the operations required by the underlying algorithms.
In our collaboration so far, we have looked into techniques for learning MTEs from data, and we have in particular pursued learning algorithms based on the maximum likelihood principle. This has resulted in a number of contributions, both at conferences and in scientific journals"