Data representation for dynamic accuracy in nuclei of neural networks

摘要

Systems for Neural Network Computation are provided. A processor of a neural network has a majority of neuronal nuclei. The processor of the neural network has one or more process organ accuracies per activation. The processor is trained to accept data with a processor frame dimension. A conversion switch is connected to the processor of the neural network and is designed to receive an input data sensor with an input accuracy per channel with one or more characteristics; to convert the input data sensor from the input accuracy to the process organ accuracy;divide the input data into a majority of blocks, each block corresponding to one of the processor core dimensions; provide each of the majority of blocks to one of the majority of neuronal cores. The processor of the neural network is designed to calculate the output of one or more layers of the neural network by the majority of neuronal nuclei.