DEALING WITH UNCERTAINTY IN A DISTRIBUTED EXPERT SYSTEM ARCHITECTURE

In this paper we describe the main characteristic of EXODUS: a distributed diagnostic expert system architecture. Knowledge in EXODUS is represented by means of a connectionist-like network and no centralized inference engine has been defined. Control is distributed among the nodes and evidence combination and propagation play a fundamental role in the reasoning process. Such an architecture is proposed as a mechanism for reaching very fastly a set of hypothesis accounting for the data characterizing a diagnostic problem (mimicing the ability of human experts who can provide initial solutions to a problem very easily). In the paper we describe in detail the architecture of the system and the mechanisms we introduced for dealing with uncertainty. In the final part of the paper we suggest that such a form of associational reasoning should be integrated with some form of deep reasoning (able to provide detailed explanations and to solve complex cases).