This project focuses on the development of novel machine architectures for communicating knowledge manipulation systems. Recent work aimed at integrating KMSs has led to the clear understanding that it is necessary to deal with knowledge data that has been produced or processed by a variety of KMSs. Thus the architectural problem is to characterise and design concurrent machines capable of coping with heterogeneous knowledge processing methods, e.g., database, rule-based, and case-based systems.

The approach we shall be taking is to make use of an architecturally-independent model of concurrency (LINDA) which is based on a type of shared associative memory called 'tuple-space'. Tuple-spaces will form the common knowledge repositories for the KMSs. An analysis of the process and communications behaviour of the LINDA-implemented KMSs (initially we are concentrating on case-based systems), will provide the foundation for deriving system architectures by giving insights into the type of machine operations that would form the instruction set of such architectures.

The unifying theme is relations. Work aimed at characterising KMSs, especially case-based reasoning systems, shows that they are mostly systems that manipulate relations, and tuple-spaces can be seen as implementations of extensional relations. It is expected therefore that the type of abstract machine operations that will be derived from the experiments with concurrent LINDA KMS implementations will be relational.

Back to the CHALK home page