An Architecture of Distributed Frame Hierarchy for Knowledge Sharing and Reuse in Computer Networks Dmitri Soshnikov Department of Numerical Mathematics and Programming, Moscow Aviation Technical University dmitri@soshnikov.com Abstract An architecture for building distributed intelligent sys- tems is considered, based on production-frame knowledge representation with distributed frame hierarchy, in which knowledge fragments located on different network nodes can be used cooperatively in the process of distributed or local inference. Frame knowledge representation can also serve as a common denominator when integrating knowl- edgebase with imperative components or relational data, because objects and component interfaces as well as rela- tional tables can be transparently represented by frames or frame classes. 1. Introduction Distributed intelligent systems can be used in many ap- plication areas. In many cases those areas can be clas- sified as distributed knowledge sharing and reuse, when distributed knowledge is applied in some manner to the solution of a problem. Such tasks occur when creating knowledge repositories for distributed knowledge-intensive virtual corporations, collections of knowledgebases main- tained by different specialists, when building large-scale ex- pert systems, etc. When using traditional multiagent approach [1] (see Fig. 1, left hand side) to distributed artificial intelligence, which is very broad and in many cases does not provide existing solution, a developer is faced with creating an application- specific architecture, models of agent communication and software tools for actual implementation. Moreover, com- plex asynchronous communication in agent society in some cases turns out to be too general, which complicates im- plementation. While in deliberative agent systems internal knowledge representation of an agent is separated from the external one that is used for communication and should be the same for the agent society, in some cases it makes sense Figure 1. Agent architecture and distributed frame hierarchy to base knowledge sharing principles on unified knowledge representation and some remote communication mecha- nisms that evolve from classical notion of remote procedure call. An approach presented in this paper is based on using production-frame knowledge representation with frame hi- erarchy distributed over computer network. Such system can be considered as one frame hierarchy that is distributed according to some remote communication protocol (see Fig. 1, right hand side), or as a collection of interoperating sub- hierarchies — in this case it can be classified as static de- liberative collaborative multiagent architecture [1] 1 . Such system can also use mobile interoperability based on inclu- sion, as opposed to static remote-call interoperability based on invocation. While for multiagent systems the structure of the prob- lem domain is typically formulated using ontologies and then translated into internal representation, presented ap- 1 Despite the fact that formally such architecture can be classified as multiagent, it can also be opposed to multiagent architecture due to the lower level of autonomy of individual subhierarchies. 1