One of the differences between software engineering and knowledge engineering is that in the latter we have to deal with incomplete and incorrect knowledge and input, and also with such complex problems that we have to fall back on heuristics, which are by definition not always correct. Reasoning in classical logic ensures a correct answer or no answer. However in the context of knowledge-based systems it is useful to also consider other forms of reasoning, which give an approximate answer under suboptimal circumstances. This allows for more gradual correctness, and can be used in anytime algorithms. An important advantage is that approximate answers can often be given more efficiently than the classical fully correct answers.
Key publications (approximate reasoning in general):
Key publications (approximate reasoning in diagnosis):
In the past decade a large number of problem solving methods have been developed. One of the current research topics is how to reuse these problem solving methods (or components of these problem solving methods) for configuring methods automatically for a specific problem. These components are taken from libraries or somewhere on the world wide web. Based on requirements on a method, the available domain-knowledge and the problem at hand, the most appropriate method should be configured. Research questions are, for instance, "how to describe the behaviour of a method", "what properties do methods have", "how to organise and index the library".
Key publications:
Currently, medical protocols are studied in both the medical domain and in AI. A medical protocol describes a procedure for particular patient groups. We study a number of KR languages for representing such protocols, and perform various forms of reasoning with such protocols (verification, critiquing, configuration etc).
Key publications:
We often use diagnostic reasoning as an example for the research themes automated configuration of knowledge-based systems and approximate reasoning.