Intelligent Systems

Research has been carried out on partitioning, mapping and implementing hierarchical VLSI neural networks. This new approach uses a divide and conquer algorithm and multi-dimensional separation planes. By taking advantage of the inherent locality in many problems, the resulting hierarchical neural network structures are much more suitable for two-dimensional VLSI implementations. The networks are developed using a more systematic approach which removes much of the "black art" of neural network implementations. The networks are ideally suited to the hierarchical connection structures which are present in almost all electronic system implementations. It has been shown both theoretically and experimentally that hierarchically connected networks map better into hierarchical hardware than globally connected networks.

Ralph Mason's Home Page

This page last updated on January 28, 1996.