Reasoning with Time Dependent Data

Abstract

Knowledge-based systems that perform monitoring and management must contend with information that changes over time. Information that is changing, is delayed and arrives out of order complicates the task of programming such systems. To simplify the construction of monitoring and management systems, a Temporal Control Structure (TCS) has been developed which manages data dependencies over time. The TCS simplifies the job of a system builder by decomposing reasoning into static and dynamic components. To the extent that the decomposition is successful, each part of the reasoning problem can be addressed separately and the overall task is easier to accomplish.

The Temporal Control Structure performs the bookkeeping tasks needed to assure that information is propagated and that the reasoning in the system is complete. Completeness means that all data entered into the system have been processed, and no more changes in the outputs of the system occur. To deal with the temporal complexities of the monitoring domain, TCS exploits two properties common in monitoring: exact knowledge of when events occur and a fixed plan for handling eventualities. These properties allow automatic scheduling of reasoning processes in response to data changes and also allow data dependencies (needed for change propagation) to be compiled into the program.

This report describes the design of the Temporal Control Structure and reports the results of several reasoning systems implemented using the formalism. The most ambitious system can track the progress of patients suffering from diabetic ketoacidosis over the course of several days. In a formal evaluation by an expert panel, the computer-generated advice was judged similar in quality to actual hospital treatment.

Massachusetts Institute of Technology, Laboratory for Computer Science technical report MIT/LCS/TR-545, August 1991.

The full paper is available in postscript. Get Postscript. (161pp)

Back to Paper List