Quantization strategy determination method of neural network and image identification method and device

Abstract

The invention relates to a quantification strategy determination method and device of a neural network, an image recognition method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring a plurality of preset quantization strategies for a target neural network; wherein each quantization strategy is used for representing the value of the operation attribute of each network layer in the target neural network, and the target neural network has corresponding model parameters under each quantization strategy; and selecting at least one quantization strategy from a plurality of quantization strategies corresponding to the target neural network as a target quantization strategy according to the test precision of the target neural networkfor testing the test sample by adopting the model parameters under different quantization strategies. By adopting the target quantification strategy determined by the method, the problem that the prediction precision of the target neural network is seriously reduced can be avoided while the target neural network is compressed.

Description

technical field [0001] The present application relates to the technical field of neural networks, in particular to a method for determining a quantitative strategy of a neural network, an image recognition method, a device, a computer device and a readable storage medium. Background technique [0002] With the development of neural network technology, neural network quantization technology has emerged, mainly for model parameter values ​​and activation values ​​(output values ​​or input values) in each network layer (such as convolutional layer and fully connected layer) in the neural network. etc., to reduce the bit width of the model parameter value and the bit width of the activation value, etc., so as to realize the purpose of compressing the data volume of the neural network model file and reducing the computing resource demand of the neural network model during the prediction process. [0003] The traditional neural network quantization technology is to directly compre...

AI Technical Summary

Problems solved by technology

[0004] However, the traditional neural network quantization technology chooses fixed bit width as the quantization strategy, which may lead to a serious decline in the prediction accuracy of the neural network

Images