A flexible, multi-level, multi-path expert database system architecture for non-deterministic rule programs
Dariush Navabi · 1991
The main objective of this research is to establish an environment which supports the specification of a wide variety of rule-based systems and facilitates fast and efficient search of the search space associated with rule programs. For this purpose the Multi-Path Architecture (MPA) is proposed. The MPA is a rule-based system with a knowledge base and multiple basic inference engines. The knowledge base consists of a representation of the rule program, the working memory contents, as well as a representation of the inference engine computational methods and data structures for the intended rule-based system. Each inference engine of the architecture can be configured using the data stored in the knowledge base to provide a virtual inference engine for the execution of the application rule program submitted to the rule-base system for execution. The MPA knowledge base is an extended relational database. Therefore this architecture is considered an expert database system. Representation of the search tree data structure and support for the instantiation process is achieved through the use of various techniques made available by the relational database technology. The MPA can also be considered as a meta-level system. Meta-level architectures refers to those inferencing systems that provide an explicit mechanism for controlling the inference process. Because of the multiplicity of configurable inference engines and explicit representation of the search tree data structure in the shared knowledge base, the architecture provides a number of advantages. It provides a convenient mechanism for combining different search methods for searching the same search space. Combination of various search methods can lead to efficient search procedures for many rule programs. Also in a multiprocessing computer system the search can be performed concurrently by multiple inference engines. The main characteristics of MPA are its flexibility, modularity, conceptual clarity and support for parallelism. The research demonstrates these characteristics of MPA and shows that the architecture belongs to the most powerful class of meta-level architectures. Finally, the meta-level overhead problem is reduced in MPA by shifting the responsibility of some of the computations involved in the inferencing process to the underlying extended relational database.