This method allows valid updates to be made quickly, efficiently, and without corruption of the existing rule base.
Mathematically sound techniques are used to view a knowledge-based system (KBS) as a set of processes executing in parallel and being enabled in response to specific rules being fired. The set of processes can be manipulated, examined, analyzed, and used in a simulation. The tool that embodies this technology may warn developers of errors in their rules, but may also highlight rules (or sets of rules) in the system that are underspecified (or overspecified) and need to be corrected for the KBS to operate as intended.
The rules embodied in a KBS specify the allowed situations, events, and/or results of the system they describe. In that sense, they provide a very abstract specification of a system. The system is implemented through the combination of the system specification together with an appropriate inference engine, independent of the algorithm used in that inference engine. Viewing the rule base as a major component of the specification, and choosing an appropriate specification notation to represent it, reveals how additional power can be derived from an approach to the knowledge-base system that involves analysis, simulation, and verification.
However, in a complex rule base that may have taken years, if not decades, to build, expecting users to have in-depth understanding of the rules that make up the system is not practical. This innovative approach requires no special knowledge of the rules, and allows a general approach where standardized analysis, verification, simulation, and model checking techniques can be applied to the KBS.
The rules of the system are likely written in a particular syntax, the possibilities for which include a language or grammar used by a particular inference engine, logic rules (written in Prolog or another logic programming or declarative programming language), propositional or predicate calculus, Horn clauses, or some form of structured English. A translator is required to translate these into the grammar of a tool that is used (and for which there is a prototype) to convert to a formal language that is process-based: that is, that recognizes that processes (or units of computation) are key components of a system.
All systems, regardless of how trivial, involve at least two processes, one being the environment in which the system is executing. Processes being enabled are analogous to rules firing; “deadlock” is equivalent to contradictions or internal inconsistencies existing in the rule base; “livelock” is equivalent to having rules that are unspecified; “top” is equivalent to overspecification; and “bottom” is equivalent to rules being underspecified. The system having been translated to the appropriate formal language, tools that already exist for that language may be applied to the analysis of the system. For example, if CSP were used as the formal language, the laws of CSP could be used to prove the absence, or otherwise, of contradictions, to pinpoint unimplemented rules, or to transform rules into a more efficient form. Then other available tools could be used for simulation and model checking.
This work was performed by Mike Hinchey, James Rash, John Erickson, and Denis Gracanin of Goddard Space Flight Center and Chris Rouff of SAIC. For more information, download the Technical Support Package (free white paper) at www.techbriefs.com/tsp under the Information Sciences category. GSC-14942-1