Learning in and for Real-Time Decision Support and Diagnosis Systems

Contents

Adaptation to changes over time as well as typical modes of usage, model calibration, and learning of preferences are some of the examples in which expert systems for real-time diagnosis and decision support benefit from learning. Complex mobile systems with these functionalities, such as those produced by DaimlerChrysler, thus require efficient real-time learning methods. Focusing on parameter and structure learning for Bayesian networks, we introduce basic realizations, illustrate their prospects and limitations, and, as the main topic of this paper, deduce the areas which will, from our point of view, most benefit from future research.