| 摘要 |
Even when the number of rules of knowledge of a knowledge base is too large in a system based on a knowledge base, bottom-up reasoning based on unit-clause reasoning in a first-order predicate logic and a rule base system (a rule base system in a broad sense inclusive of a production system) can be sped up as a whole by speeding up the selection of rules used in bottom-up reasoning by parallel processing. The structure in which each process unit in each processor is divided by every scalar necessary for the number of kinds or by every logical value functions in such a manner that each of all the scalars or all the logical values after division is processed in parallel. When this structure is utilized for addition/subtraction of number-of-kind vectors, parallel processing can be improved. When the structure in which the rules of knowledge are divided into modules each grouped by a local predicate is utilized, or when the structure in which each module is divided into small modules to which at least 32 rules belong in such a way that the facts of a unit clause necessary for each small module can be reduced, the number of times of parallel processing can be reduced. Particularly, speed-up by parallelization in a single processor can be accomplished.
|
