Implementation of Composition-Based-Learning Production Systems and Characteristics
Project/Area Number |
01550296
|
Research Category |
Grant-in-Aid for General Scientific Research (C)
|
Allocation Type | Single-year Grants |
Research Field |
計算機工学
|
Research Institution | Fukuoka Institute of Technology |
Principal Investigator |
ARAYA Shinji Fukuoka Inst. of Tech., Professor, 工学部, 教授 (60175974)
|
Project Period (FY) |
1989 – 1990
|
Project Status |
Completed (Fiscal Year 1990)
|
Budget Amount *help |
¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 1990: ¥400,000 (Direct Cost: ¥400,000)
Fiscal Year 1989: ¥900,000 (Direct Cost: ¥900,000)
|
Keywords | macro-operator / EBL / chunking / rule composition / rule decomposition / pattern matching / Rete algorithm / knowledge transformation / 解析的学習 / プロダクションシステム / 学習 / システム性能 |
Research Abstract |
1) A lot of methods that improve the performace of deductive reasoning systems are discussed and compared each other being classified into three levels : interpretation/execution level, pattern matching level and knowledge level. 2) An interactive composition-based learning production system was implemented by adding a rule-composition mechanism to the Common Lisp version of OPS5. 3) A knowledge filter has been developed, which automatically evaluates composed rules and deletes ones that does not improve the performance. 4) It is shown that the rule composition under state-saving pattern matching algorithms could generate expensive rules and increase both space and time for the matching. 5) The rule composition and the partial matching can be used together. Three kinds of integration methods of the two mechanisms suitable for the characteristics of problem domains are proposed. 6) The qualitative analysis has clarified that three algorithms for pattern matching : non-saving, partial saving (Treat) and complete saving (Rete) have different problem domains where they can exhibit their full power. 7) In order to resolve the expensive rule problems, we proposed a selective join mechanism which caluculates instantiations of expensive rules little by little when they are required. 8) We proposed a new knowledge transformation called "rule decomposition" that decomposes one expensive rule into suitable size of sub-rules. An analysis and experiments have shown that rule decomposition can greatly improve both time and space efficiency.
|
Report
(3 results)
Research Products
(19 results)