Semantic web and grid technologies offer a promising approach to facilitate semantic information retrieval based on heterogeneous document repositories. In this paper the authors describe the design and implementation of an Ontology Server (OS) component to be used in a distributed contents management grid system. Such a system could be used to build collection document repositories, mutually interoperable at the semantic level. From the contents point of view, the distributed system is built as a collection of multimedia documents repository nodes glued together by an OS. A set of methodologies and tools to organize the knowledge space around the notion of contents community is developed, where each content provider will publish a set of ontologies to collect metadata information organized and published through a knowledge community, built on top of the OS. These methodologies were deployed while setting up a prototype to connect about 20 museums in the city of Naples (Italy).

This is the first of two articles presenting an approach to rule-based expert systems for diagnostic tasks exploiting a purely neural architecture. Here, we outline the methodological options motivating this approach, and describe a forward and backward chaining mechanism on a system of production rules. This inference engine is furnished with an informative justification module, which exploits the fact that most individual neurons get a precise semantic assignment in terms of the literals appearing in production rules. The control and synchronization functions needed to schedule these processes are carried out by a neural network, too. 0 1995 John Wiley & Sons, Inc. I. PURELY NEURAL ARCHITECTURESWhich parts of expert knowledge and reasoning can be adequately and efjciently captured by neural networks, and by which types of such networks? This is the pertinent question to be addressed when exploring the possible uses of neural elements in the design of a (diagnostic) expert system. A standard answer to this question is that neural nets are unsuitable for stepwise expert reasoning, possibly involving subgoaling and interactions with the environment, as well as for the related task of providing informative explanation facilities. In contrast, the learning and generalization capacities of neural nets can be profitably employed, e.g., for handling noisy or incomplete data, acquiring expert knowledge by examples, solving one-step classification problems. '-6 This popular answer is partially misleading, because many types of nonconnectionist networks can be used to carry out elaborate forms of stepwise reasoning (including justification). Indeed, representing the knowledge bases of most diagnostic expert systems does not require an essential use of a first-order language, for these are supposed to capture relations over a finite domain of elements such as components and (ab)normal states of a class of objects, and explanatory hypotheses for the observable anomalies. In turn, such knowledge bases can be codified by means of (weighted) production rules over propositional

No abstract

No abstract