Neural network quantification method and system

Abstract

The embodiment of the invention provides a neural network quantization method and a system, and the method comprises the steps: (1) obtaining neural network parameters, carrying out the fusion of a convolution layer and a BN layer in the forward propagation of an original network, so as to obtain a new to-be-quantized weight and a to-be-quantized bias, and storing the activation output data of each layer at the same time; (2) quantizing neural network parameters and data distribution obtained after fusion; and (3) forming a new quantization network by taking the quantization parameter obtainedafter quantization processing as a parameter of forward propagation of the quantized new network. According to the method, various problems caused by too large or too small quantization scaling scaleof the neural network can be effectively balanced, and the operation speed is increased while hardware resources are saved.

Description

Technical field [0001] The invention relates to the field of neural networks, and in particular to a neural network quantification method and system. Background technique [0002] Neural network is one of the core technologies that have made significant progress in artificial intelligence technology in recent years. With the continuous increase in computing power, the rapid development of the Internet and the Internet of Things has also provided massive amounts of data. For this reason, people have developed various Complicated neural network calculation methods and models have made remarkable progress in intelligence. [0003] As the algorithm becomes more and more complex, the requirements for the computing power of the device are also higher and higher. For some scenarios, such as mobile terminals, wearable devices, offline computing, edge computing and other scenarios, the intelligentization brings more A big challenge. Therefore, for these scenarios, the quantification or acc...

AI Technical Summary

Problems solved by technology

[0006] However, these existing technologies mainly focus on how to quantify the weight parameters and input features of the convolutional neural network, without considering the impact on the calculation process of the original convolutional neural network when it is implemented on hardware after quantization, which leads to differences with the original There is a certain bias in the neural network, which reduces the robustness of the convolutional neural network

Images