Connectionist Rule Processing Using Recursive Auto-Associative Memory

摘要

A limitation of many rule-based connectionist models is their dependence on structure to explicitly represent rules, and their consequent inflexibility in acquiring and applying novel rules. A model is described in which recursive auto-associative memory (RAAM) is used as an encoding mechanism to prepare rules of variable structure and content for input to a connectionist rule applicator. The encoded rules are applied 'holistically' to perform simple operations upon binary strings.