Approximations in diagnosis: motivations and techniques
Frank van Harmelen, Annette ten Teije · 1995
We argue that diagnosis should not be seen as solving a problem with a unique definition, but rather that there exists a whole space of reasonable notions of diagnosis. These notions can be seen as mutual approximations. We present a number of reasons for choosing among different notions of diagnosis. We also present an exhaustive categorisation of techniques that can be employed to obtain approximations, as well as a number of specific example techniques for each category. We also show that it is possible to characterise the relations between the approximations obtained by these techniques. 1 Introduction The AI literature contains many definitions for diagnostic reasoning. Implicitly or explicitly, these papers make many assumptions about the kind of knowledge that is assumed to be available, and the kind of properties that a diagnosis is supposed to satisfy. Tacitly these papers claim that if these assumptions are satisfied, they present the appropriate definition of diagnosis. Ho...