The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #92140

Title:A Connectionist Representation of Rules, Facts, and Dynamic Bindings of Variables
Date: 1992
Abstract:The ability of humans to reason with bast amounts of structured knowledge has been considered a main characteristic of cognition. When humans perform language understanding, text understanding and commonsense reasoning, they do relevant reasoning without any conscious efforts. In a recent technical report, Shastri and Ajjanagadde proposed a partial connectionist model for such reasoning which requires very fast inference. The described in detail a technique for using neuron-like elements to encode facts and rules involving n-ary predicates with variables. In their model, they adept temporal synchrony to achieve dynamic binding of variables which has been one of the weaknesses of connectionist systems. A key part of their mechanism to solve the dynamic variable binding problem is in matching rhythmic patterns of activity, i.e. temporal synchrony, within the network. This thesis describes the exploration of the proposed connectionist mechanism to answer the following questions: How are the n-ary predicates, rules, and facts represented in a connectionist manner for structured knowledge representation ? What is the core mechanism of temporal synchrony to solve the dynamic variable binding problem ? How biological plausible and efficient is the inference achieved ? What are the constraints of the proposed mechanism ? During the course of the project an implementation of the proposed connectionist rule-based system and clarification of the constraints on the proposed mechanism has been achieved.

[Search These Pages] [DAI Home Page] [Comment]