Neural network quantification method and device and neural network reasoning method and system

The invention relates to a neural network quantification method and device and a neural network reasoning method and system in the technical field of information processing. The method comprises the following steps: setting a reasoning precision threshold value of a neural network; separating and quantizing each layer of output feature data quantization parameter and each layer of weight quantization parameter; for the front and back associated layers in the neural network, updating the output feature data quantization parameter of the input layer according to the output feature data quantization parameter of the current layer; and by taking the reasoning precision threshold as a constraint condition, according to a comparison result of each layer of reasoning floating point value and the separated and quantized fixed point value, optimizing and outputting a characteristic data quantization parameter. And executing a fixed-point reasoning process of each network layer by using a coprocessor in the Soc chip. A