A connectionist architecture, dubbed RUBICON, for implementing rule-based systems is described. RUBICON uses both distributed and local representations. Input and output are fully distributed, allowing the use of microfeatures for robust interfacing to the external world. All input units, however, are local. The local internal representation results in straightforward network definition for a set of rules, and it should also facilitate adaptation. There are a number of features that distinguish RUBICON from previous connectionist attempts at developing rule-based systems. These include: an arbitrary number of antecedent and consequent expressions in any rule, the addition and deletion of working memory elements, chained inferencing, and the handling of negated expressions in antecedent and consequent clauses. A toy expert system for film identification is used as an illustration.
|Original language||English (US)|
|Title of host publication||IEEE Int Conf on Neural Networks|
|Publisher||Publ by IEEE|
|Number of pages||8|
|State||Published - Dec 1 1988|