A wavefront `snake' architecture for multilayer neural networks

摘要

Summary form only given, as follows. A digital wavefrontarchitecture implementing a multilayer neural network based on thebackpropagation learning algorithm is presented. The proposedarchitecture is a linear array of locally interconnected elementaryprocessors which resembles the form of a snake if implemented folded ona plane. This architecture has several advantages which make it veryflexible: it has only local connections; it can be expanded by simplyadjoining more processors; it can be configured in terms of number andwidth of layers; it permits pipelining the data to be processed. In theforward mode it is able to reach up to 100% efficiency