| Reasoning under uncertainty |
|
Organization responsible: TUD People involved:
Project description: Decision support systems are often required to operate on-line with incomplete knowledge about their environment, which may be changing in an unpredictable manner. In order to cope with these changes and to learn from past experience, these systems have to actively acquire and organize knowledge about the surrounding world, to plan future subgoals and actions toward the achievement of these subgoals. However, in real-world situations, one has to cope with the uncertainty and even partially contradicting information in data, e.g., when the data are obtained through different input channels (sensors). Uncertainty in the currently available knowledge (world model) has to be considered as well. The relevance of the knowledge may depend on the particular situation. Therefore, methods must be developed for intelligent filtering of information, and for the possible reduction of uncertainty. In order to develop an effective framework for data fusion and interpretation under uncertainty, several techniques were employed. First, suitable forms of uncertainty representation were investigated in order to create a common frame in which data and knowledge can seamlessly be combined and updated. In order to keep this learning process manageable, the system must be able to control the complexity of the data and knowledge. Therefore, automated reduction / simplification tools needed to be applied in order to manage the complexity. An important class of systems can be represented as cascaded, or, in general, distributed subsystems. However, the distributed approach, in general, does not guarantee the same performance as a centralized one. Tools have been developed in order to keep the system tractable while satisfying imposed performance criteria. Emphasis will be put on robust performance of the system under varying conditions. Publications: |