processing and computer simulation of a neural network structure for this neural network structure

摘要

The invention relates to a method of calculating norms, implemented in a neural network structure or a processor. This enables the norm of the new synaptic coefficients to be determined without requiring explicit calculation thereof. For this purpose, the structure comprises a processing unit (10) which determines output potentials yk and increments DELTA k, a unit (13) for storing the old and new norms, and a unit (15) for calculating the new norms. The latter are thus determined rapidly without requiring the calling up of the synaptic coefficients in memory. The method can be used with certain algorithms which require a calculation of norms. Application: neural networks. IMAGE