| Abstract
|
We formally describe a framework that we s … We formally describe a framework that we suggested previously for abstracting time-stamped data and sharing the domain-specific knowledge needed for this task. The temporal-abstraction (TA) task is an important part of general tasks such as pattern recognition over time and planning (e.g., forming and revising therapy plans for patients). The TA task is decomposed by the suggested TA method into several subtasks achieved by three basic TA mechanisms: point temporal abstraction, for abstracting values of several temporally coexistent parameter values into a value of another parameter; temporal inference, for inferring domain-specific sound logical conclusions for a single temporal interval or two meeting intervals; and temporal interpolation, for bridging nonmeeting temporal intervals. The TA method has also been implemented as a system, RESUME. We then show how this framework can be formally specified as inferences in the KADS knowledge-modeling methodology, thus proposing a platform for modeling and sharing domain-independent TA inference knowledge. Making explicit the knowledge required for TA and formalizing the semantics of the mechanisms involved supports the acquisition of that knowledge by automatically generated tools and emphasizes the advantages of a modular, task-specific but domain-independent architecture for designing knowledge-based systems. ure for designing knowledge-based systems.
|
