Concurrent Computer Corp is hitting the expert systems market with a version of Donald Michie’s Extran running in close conjunction with its own Reliance transaction processing database system. While marketing manager David Steel is anxious not to give the impression that there is a product ready for sale, the company has brought together Professor Michie’s Intelligent Terminals Ltd and its own team of database developers in the UK, and will release the relevant interface to users who have the type of application that can benefit from it. It will be called Reliance Expert and has been used by US consultant Ken Moddesit to build a system to analyse dada from the Space Shuttle project. The Shuttle project produces reams of graphs and figures which can reveal problems such as those leading to last year’s disaster, but the volume of data is so vast and the number of expert who can make any sense from it so rare that an expert system was the only way forward. Moddesit has been a party to 1,400 tests on the Shuttle project, each producing 50 megabytes of data. Extracting rules from all this data may have taken quite a while, so an inductive system such as Extran, which induces its own rules from a knowledge base would have been ideal if it could get access to all the existing data. Steel added that it was also felt that a system that induces its own rules from examples was less likely to induce an incorrect rule from that mass of data than a system where you have to isolate and define the rules before entering them. A division of Rockwell International carried out the test on the Shuttle but needed extra coding that would allow examples to be automatically extracted from Reliance from which Extran could induce its rules. The system asks for selection criteria and the relevant attributes of the data you want from the Reliance database to build examples with; after that the process is all automatic. This is in essence what Reliance Expert is and after a handful of Concurrent customers get some experience with it, should form the basis of a general release product. Steel adds We are working on a version that can go away at run time and take its rules from current data rather than old data, and we see no reason why it couldn’t be adapted to real-time applications, especially since Extran rules are generated in Fortran, which is favoured in many real-time environments.One of the major criticism of inductive systems is that you often don’t know what rules are being applied and whether or not they make sense, but Extran can put the rules it applies on screen along with a decision tree.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.