Using contextual information in decision making
Résumé
Look-ahead reasoning leads to unmanageable decision trees due to the large number of possible actions and events. However, in decision practice, people reduce the complexity of the tree by using as much as contextual information as they can. In this paper, we explain and model the different ways of using contextual knowledge to reduce decision-tree complexity. From a theoretical viewpoint, we introduce first the notion of action robustness and action postponing. As a consequence, the reasoning moves from look-ahead to diagnosis and relies on macro-actions. Then, we illustrate how context can be modeled and used to simplify decision trees according to the notions introduced in the first part.