Information Modeling: Extending the model concept
Condensing information into models is a key feature in science, and it takes different forms in different disciplines. In physicsand systems engineering the basic format is difference/differential equations, constructed either from measured data or from physical laws black box and grey boxmodels. In Artificial Intelligence the dominating models arelogic, rule based collections of statements, or qualitative classifications. For visualization, e.g., morphogenetic models describe shapes and structures. In advanced applications, it willbe fruitful to merge ideas from different areas to develop richer integrated model concepts. Some specific research problems are:
- Hybrid Models are models that merge differential equations and logical statements. This is an area in its infancy, but of crucial importance for integration of model-based deliberative, reactive and control capacity. An early approach exists, and methods to estimate such models have been outlined. On-line architectural support for the use of hybrid models in complex systems is equally crucial but surprisingly absent from many research agendas.
- Black and Grey Box Models: Merging techniques from physical modeling and data-based modeling. Physical modeling with state-of the art techniques like Modelica leads to systems of differential-algebraic equations. It is an important challenge to combine these modeling tools with statistical methods for model estimation.
- Morphogenisis and Data Structures: A fascinating idea is to use new techniques in self-organized geometric modeling for the urgent need of representations of complex multidimensional data. It would be an important complement to traditional statistical techniques, to find insightful projections of observed data. This could be applicable in areas as diverse as visual data mining of large data warehouses and for grey box models for complex physical phenomena such as tissue representation for haptic interaction in medical applications.
- Model Integration: Integrating models from different areas suchas standard differential equations with those of geographicalinformation systems (GIS), and models used for situationawareness will be needed in future vehicle safety systems andautonomous vehicles. SLAM (Simultaneous Localization AndMapping) is a typical example of building complex models fromdata bases and measurements. Particle filters offer one promisingtool for merging such diverse models and making inferences fromobservations combined with such models.