Neural network quantitative classification method and system

Abstract

The invention provides a neural network quantitative classification method and system. The neural network quantitative classification method and system are different from the prior art in which only the weight of the neural network model is classified and quantized. According to the method and the system, statistical processing, classification processing and weight quantification processing are carried out on the input information of each layer in the neural network model, so that proper weight quantification processing can be carried out on the output information of each layer in a targeted manner; therefore, the precision of the output information of each layer is improved to the maximum extent under the same quantization bit number. According to the method and the system, the output information of each layer of the neural network model is sorted; in this way, a proper calculation mode can be selected according to the respective output channel characteristics of the output information of each layer, so that the neural network model can obtain the optimal calculation efficiency and reduce the power consumption required by hardware operation during hardware operation.

Description

technical field [0001] The present invention relates to the technical field of neural networks, in particular to a neural network quantitative classification method and system. Background technique [0002] The purpose of quantifying the weights of the neural network model is to classify the weights of the neural network model and share the weights in each category under the premise of ensuring that the performance of the neural network model for the target task does not decrease significantly, so that The effect of reducing the storage space of the neural network model is achieved. Specifically, the weights of each different stage of the neural network model are classified, and different quantization weights are used between different classes, which can effectively reduce the loss of precision caused by quantization, and perform weights on the weights of the neural network model The above classification process can use as few bits as possible to quantify the weights, so as...

AI Technical Summary

Problems solved by technology

[0003] However, the prior art only classifies the weights of the neural network model and quantizes different bits, and does not optimize the quantization of the input information of each layer in the neural network model, which seriously hinders the neural network model. Further Improvement of Network Model Computational Efficiency

Images