Knowledge Base Compilation (1986-1990)*
... Efficient Implementation of Knowledge Based Expert Systems
Knowledge Base Compilation is a technique for increasing
performance and portability of expert systems. It was developed at the
University of Kiel (CAU)
. Knowledge Base Compilation makes expert system
applications efficiently available even on small computers within
industrial application environments. The work was done within a
project sponsored by the Dr.-Ing. Rudolf Hell GmbH, Kiel, and in
cooperation with the Expert System
Research Group of the German National Research Center for
Computer Science ( GMD ),
St. Augustin, Germany. At Kiel we developed new compiler
techniques for hybrid knowledge representation languages, and a
concrete compiler for BABYLON which
was used successfully to compile expert system applications of industrial
relevance onto PC AT's and even notebook computers.
- BCLiC: The BABYLON to Common Lisp Compiler.
- CLiCC: The Common Lisp to C Compiler.
- BABYLON : AI Workbench.
BABYLON knowledge bases are compiled to Common Lisp in a first step
using the BABYLON to Common Lisp Compiler BCLiC. In a second step, the resulting program may
be further compiled to C using the Common Lisp to C Compiler
Compilation increases portability. But moreover, we made compiled
expert systems run about 10 to 20 times faster than their interpreted
counterparts running under BABYLON. Compiled expert system
applications run on PC's just as fast as the interpreted versions on
large workstations. Using the implementation technique
Knowledge Base Compilation we obtain
Modern tools which allow incremental program development and rapid
prototyping with adequate knowledge representation languages are
useful for building expert systems. Our work shows that expert system
applications can be made efficiently available as stand alone
applications on a large variety of application machines without
the need for expensive reimplementation.
- portability of expert systems,
- performance enhancement of expert systems, and
- encoding of expert knowledge within expert systems.
* The work has been supported by the Dr.-Ing. Rudolf Hell GmbH, Kiel.