Diagnostic Improvement through Qualitative Sensitivity Analysis and Aggregation

Keith L. Downing

This paper lays the foundation for a diagnostic system that improves its performance by deriving symptom-fault associations from an underlying causal model and then utilizes those relationships to impose further structure upon the "deep" model. A qualitative version of sensitivity analysis is introduced to extract the implicit symptom-fault information from a set of local constraints. Parameter aggregation triggered by this new information then simplifies diagnosis by forming a more abstract causal representation. The resulting diagnostician thus employs both an experiential and a first-principle approach, where in this case "experiences" are compiled directly from first-principles. Key issues include the roles of knowledge compilation and abstraction in refining qualitative models of physical systems.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.