Quantitative calculation method and system for convolutional neural network

Abstract

The invention relates to the field of neural network algorithm hardware implementation, and discloses a quantitative calculation method and system for a convolutional neural network. The quantitativecalculation method comprises the steps of: allowing all calculation layers of a convolutional neural network to be respectively matched and quantized in a multi-valued quantification mode and a multi-bit quantification mode according to the calculation precision and calculation capability requirements, allowing the calculation layers after multi-bit quantification to be mapped to a high-precisionarray, and carrying out high-precision calculation; and mapping the calculation layers after multi-bit quantification to a high-calculation-power array, performing high-calculation-power calculation,and completing calculation of the convolutional neural network according to a high-precision calculation result and a high-calculation-power calculation result in combination with non-calculation layers. According to the invention, the reasoning speed of the convolutional neural network is increased; the accuracy is ensured; meanwhile, the network power consumption is reduced as much as possible;and high practical value and wide application prospect are achieved.

Description

technical field [0001] The invention relates to the technical field of neural network algorithm hardware implementation, in particular to a convolutional neural network quantization calculation method and system. Background technique [0002] Convolutional neural networks have shown great advantages in image recognition, object detection, and many machine learning applications. The convolutional neural network is mainly composed of a convolutional layer, a pooling layer, and a fully connected layer cascade. It mainly has the following operations, namely, the convolution operation between the pixel block and the convolution kernel, the activation operation for introducing nonlinearity, The downsampling operation (i.e., pooling) and full connection operation on the feature map to reduce the feature value. Among them, most of the calculations are in the convolutional layer and the fully connected layer. [0003] Large-scale convolutional neural networks have huge parameter se...

AI Technical Summary

Problems solved by technology

In the existing NOR Flash-based storage and computing integrated computing array, some control circuits are complex, resulting in high power consumption and insufficient computing power; some Flash unit threshold voltages are only divided into high and low binary values, and the corresponding control circuits are simple and easy to construct. A large-scale Flash computing array with low computing power and low power consumption, but the calculation accuracy is low and the accuracy rate is not high

Images