On efficient matching and optimization of rule sets in AI and databases
Expert systems with large knowledge bases have necessitated the use of database technology for efficient knowledge management. Emerging applications like process control, battle management, and office automation require rule-based reasoning on a large disk-based database. Forward chaining production...
Saved in:
| Main Author: | |
|---|---|
| Format: | Dissertation |
| Language: | English |
| Published: |
ProQuest Dissertations & Theses
01-01-1990
|
| Subjects: | |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Expert systems with large knowledge bases have necessitated the use of database technology for efficient knowledge management. Emerging applications like process control, battle management, and office automation require rule-based reasoning on a large disk-based database. Forward chaining production systems, e.g. OPS5, are commonly used to implement expert systems. The close match between the data model of the OPS5 production language and the relational model makes their integration highly desirable. However, existing techniques in both domains are unsuitable for a successful integration. A production system consists of rules, defined as $\langle condition,\ action\rangle$ pairs, that repeatedly performs a three-phase match-select-execute cycle. Each rule is matched against the data stored in a global database (working memory) to determine whether its condition is satisfied. If so, the rule is inserted into the conflict set. One rule is selected in the select phase for firing in the execution phase. Matching is a time consuming activity even in a main memory environment whose degraded performance is further aggravated in secondary memory. This thesis focuses on both the static and dynamic aspects of matching. The static aspect concerns the construction of an efficient matching network while the dynamic nature addresses efficient matching. It is known that the static aspect is NP-Hard and is also important in other domains. We will show that there exists an equivalence in matching problems in other domains, e.g., the view update problem, multi-query optimization and integrity maintenance in database systems, and that a solution in one domain is also applicable to the other domains. The focus on the dynamic aspect is on the design of efficient algorithms for matching due to the bottleneck in a disk-based environment. Typical matching algorithms are either tuple space based or binding space based. Tuple space based matchings are space efficient at the expense of slow convergence while binding space based ones have fast convergence but are space inefficient. We propose a hybrid space matching algorithm that yields fast convergence and is space efficient. Our matching algorithm is also applicable in the other domains as well. We also provide simulation results to support our observations. |
|---|---|
| ISBN: | 9798207024592 |