Date of Award
Doctor of Philosophy (PhD)
W. Pratt Mounfield
A novel approach using Neural Networks has been developed to generate consistent labelling of facts in relation to a given set of rules. In the proposed system, facts are represented by neurons and their interconnections form the knowledge base. The Neural Network Truth Maintenance System(TMS) arrives at a valid solution provided the solution exists. A valid solution is a consistent labelling of facts. If a valid solution does not exist the network does not converge. An experimental setup was built and tested using conventional integrated circuits. The hardware design is suitable for VLSI implementation for large, real-time problems. The TMS Neural Network blends the simplicity and speed of Neural Network architecture with the power of artificial intelligence techniques. A methodology has been developed to study the stability of logical networks in terms of Lyapunov Stability criteria.
Guddanti, Suresh, "A Neural Network Truth Maintenance System." (1991). LSU Historical Dissertations and Theses. 5185.