If we wish to replicate symbolic styles of inference in a connectionist system we are constrained to work with simple network elements, with localised behaviours and simple forms of link to others. However, there remains considerable scope for experiment with different forms of element and links. This paper motivates our choice of a particular element and link arrangement which we believe to be more uniform in structure than other related architectures while also allowing a closer match to the basic elements of symbolic inference which we wish to explore. Using this element, this paper describes how dynamic bindings and basic symbolic inference procedures (matching and substitution) are represented in a connectionist manner. The paper also provides entry points to more extensive descriptions of the system and its use.
|Title of host publication||Proceedings of the ECAI-96 Workshop on Neural Networks and Structured Knowledge|
|Number of pages||7|
|Publication status||Published - 1 Aug 1996|