An Architecture for the Integration of Heterogenous Inference Systems

Embedded approaches to the design of hybrid architectures for knowledge representation and reasoning organize the component inference tools around a designated kernel, which provides the basic problem-solving mode of the architecture. This approach suffers from lack of flexibility as the chosen kernel might not always provide the most appropriate control and problem-solving architecture for the task in hand. Moreover, because embedded approaches emphasize communication between the component tools and the kernel, they typically provide only limited mechanisms for cooperation and communication between component tools. An alternative method of integrating inference systems is through the design of nonembedded architectures, where the component inference tools cooperate as 'equals', as none of them provides the structuring problem-solving organization. This alternative approach is more flexible and modular, and emphasizes aspects dealing with inter-tool communication and cooperation. However, as tools in non-embedded architectures are more loosely integrated, it is more difficult to ensure the coherence and consistency of the overall hybrid knowledge base. In this paper we describe a non-embedded, open architecture for knowledge representation (VITAL-KR), which provides a framework for integrating assimilable, heterogeneous representation and reasoning systems. The aim of the VITAL-KR is to provide a flexible and modular programming environment in which it is possible to devise specific hybrid configurations, in response to specific application needs. The component inference tools are fully integrated, as they can share, communicate, and request knowledge from other tools. Moreover, a mechanism for integrating local (to a particular tool) and global (to the whole hybrid knowledge base) truth maintenance techniques is provided. In this way, global consistency can be maintained without loss of efficiency. An Architecture for the Integration of Heterogeneous Inference Systems Page 1 © The Open University 1991