Probabilistic Similarity Networks (ACM Doctoral Dissertation Award)

In this remarkable blend of formal theory and practical application, David Heckerman develops methods for building normative expert systems expert systems that encode knowledge in a decision-theoretic framework.

Heckerman introduces the similarity network and partition, two extensions to the influence diagram representation. He uses the new representations to construct Pathfinder, a large, normative expert system for the diagnosis of lymph-node diseases. Heckerman shows that such expert systems can be built efficiently, and that the use of a normative theory as the framework for representing knowledge can dramatically improve the quality of expertise that is delivered to the user. He concludes with a formal evaluation of the power of his methods for building normative expert systems. David Heckerman is Assistant Professor of Computer Science at the University of Southern California. He received his doctoral degree in Medical Information Sciences from Stanford University.

Contents: Introduction. Similarity Networks and Partitions: A Simple Example. Theory of Similarity Networks. Pathfinder: A Case Study. An Evaluation of Pathfinder. Conclusions and Future Work.