Parallelization Of Inference Systems By Using An Abstract Machine PDF Download

Abstract: "In this paper we will present a framework for the sequential and parallel execution of a variety of different inference systems. For the execution of an inference system, several basic building blocks can be identified (see [Wil90]): Input Language, Basic Data Types, Structure Building, Parameter Transfer, Searching, Side Effects, and Stop Condition. By selecting an appropriate value for each building block, the execution of the inference system can be determined. For the execution of inference systems, we define an abstract machine, the AIM, which has machine instructions for each of the possible values of each building block. By selecting the appropriate set of machine instructions, an interpreter (or code templates for compilation) can be written, using these instructions. This approach also facilitates the parallelization of the given inference system: for each of the building blocks, several different kinds of parallel model can be detected. Since each building block represents a number of AIM instructions, parallelization can be accomplished easily by exchanging these instructions by their parallel counter-parts. The other instructions are left untouched. Some examples will be given to illustrate this approach."