A digital implementation of a multilayer neural network model that has backpropagation as its learning algorithm is presented. This architecture is characterized by a set of elementary processors (neurons) and has the form of a linear sequence, where every processor communicates only with its two nearest neighbors. A sophisticated control of data exchange among neurons, by means of two data buses, ensures full pipelining in forward mode. The proposed architecture is very flexible since, having only local connections, it can be easily expanded by simply adjoining more processors to it. Moreover, it can be programmed in terms of number and width of layers.
A Digital "Snake" Implementation of the Back-Propagation Neural Network
MARCHESI, MICHELE;
1989-01-01
Abstract
A digital implementation of a multilayer neural network model that has backpropagation as its learning algorithm is presented. This architecture is characterized by a set of elementary processors (neurons) and has the form of a linear sequence, where every processor communicates only with its two nearest neighbors. A sophisticated control of data exchange among neurons, by means of two data buses, ensures full pipelining in forward mode. The proposed architecture is very flexible since, having only local connections, it can be easily expanded by simply adjoining more processors to it. Moreover, it can be programmed in terms of number and width of layers.
I metadati presenti in IRIS UNICA sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono protetti da diritto d'autore, salvo diversa indicazione.
