Convolutional neural network calculation optimization method and device, computer device and medium

Abstract

The embodiment of the invention discloses a convolutional neural network calculation optimization method and device, a computer device and a medium. The method comprises the steps of obtaining a to-be-input feature map of an optimized convolutional neural network; determining a matched input adjustment mode according to the relationship between the optimal size of the feature map corresponding tothe optimized convolutional neural network and the size of the to-be-input feature map, and adjusting the size of the to-be-input feature map; inputting the adjusted feature map to be input into the optimized convolutional neural network to obtain an output feature map output by the optimized convolutional neural network; and determining an output adjustment mode of the output feature map according to the input adjustment mode, adjusting the output feature map, and taking the adjusted output feature map as a target output feature map result of the convolutional neural network. According to theembodiment of the invention, the convolutional neural network can adapt to the acceleration performance of hardware equipment, and the calculation speed of the convolutional neural network is improved.

Description

technical field [0001] The embodiments of the present invention relate to the field of data processing, and in particular to a convolutional neural network calculation optimization method, device, computer equipment and media. Background technique [0002] With the rapid development of big data, the demand for neural network models is increasing. Usually, the neural network model needs to be trained by sample data to form a model with preset functions, so as to perform data processing operations such as prediction of unknown samples. [0003] At present, as the amount of data increases, the computational load of the neural network model becomes larger and larger, resulting in low computational efficiency of the neural network model. In order to solve the above-mentioned problems, an existing method is to use hardware to accelerate. For example, in a convolutional neural network model, hardware circuits, such as Tensor Core, can be designed for convolution to speed up the o...

AI Technical Summary

Problems solved by technology

[0004] However, based on the consideration of reducing chip area and power consumption and reducing the complexity of hardware design, the design of Tensor Core will only provide extreme acceleration for some specific convolution shapes (rows and columns of convolution kernels), and for some that do not meet Convolution of a specific shape, the acceleration performance of Tensor Core is poor

If the trained parameters of the convolutional neural network model are changed to adapt to the Tensor Core, the convolutional neural network will need to be retrained, which will increase the computational load of the convolutional neural network and reduce the efficiency of the convolutional neural network.

Images