Neural network for dynamic processes

摘要

The invention concerns a neural network with a layer of processor elements, in which network input magnitudes which are switched to one input of the network are switched to each processor element with weights depending on an input matrix (G). Each processor element has an activity which depends on a sum of weighted input magnitudes, and supplies an output magnitude which is a function of the activity. The output magnitudes of the processor elements are superimposed, weighted depending on an output matrix (H), to form output magnitudes of the network at an output. The input magnitudes of the network are switched to the input clocked at a specified clock rate. Also, weighted output magnitudes of all processor elements are applied to the input of each processor element. The activities of the processor elements in a clock cycle (k + 1) are specified by the weighted input magnitudes and weighted output magnitudes in the previous clock cycle (k). <IMAGE>