Reduction of Diagnostic Complexity through Model Abstractions

ions Igor Mozetic Austrian Research Institute for Artificial Intelligence Schottengasse 3, A-1010 Vienna Austria igor@ai.univie.ac.at Abstract The paper presents an application of abstractions to model-based reasoning with the goal to improve the efficiency of diagnosis. We define three abstraction operators which map a detailed, complex model to an abstract, simpler one. Our approach is not restricted to discrete, qualitative models. We define a provably correct and complete diagnostic algorithm and analyse its complexity. Abstractions and the algorithm were applied to a complex medical problem: ECG interpretation based on the model of the heart's electrical activity. Results show that hierarchical diagnosis has logarithmic time complexity when compared to a one-level diagnosis --- in particular, a speedup of a factor of 20 was achieved. 1 Introduction There are two fundamentaly different approaches to diagnostic reasoning. In the first, heuristic approach, one encodes diagnostic ..