Selection algorithms for models with context specific independencies

Graphical models have become a popular tool for the analysis of categorical data, where the possibility to interpretate the model in terms of conditional independence is of particular importance. It is natural to extend these models to situations, where conditional independence holds only for specific configurations of the conditioning variables. This model class is much larger, and consequently there is a strong need for doing efficient model selection. The present project deals with forward/backward model selection algorithms, where the search space is limited to models, which can be decomposed. This means, that the search can be based on local and exact calculations. Besides, the models can be illustrated by a marked graph, which is a completely natural extension of the graph associated with a graphical model.