Computing resource optimization method and system for convolutional neural network

Abstract

The invention discloses a calculation resource optimization method of a convolutional neural network. The calculation resource optimization method of the convolutional neural network comprises the following steps: splitting an input Map of the convolutional neural network into a sub-Map matrix including several sub-Maps , wherein, the size of the input Map is H×W, H>0, W>0; a separate convolution operation is performed on each of the sub-Map to obtain the calculation result of the convolution operation of each of the sub-Map ; According to the calculation results of the convolution operations of all the sub-Maps, the calculation results of the convolution operations of the input Map are spliced ​​in situ, wherein the in-situ splicing refers to placing the results of the convolution operations of the sub-Maps in the The corresponding position of the sub-Map in the input Map. The invention also discloses a calculation resource optimization system of convolutional neural network. The technical scheme of the invention can achieve the goals of saving computing resources and improving computing real-time performance.

Description

technical field [0001] The present invention relates to the technical fields of Machine Learning (ML) and Artificial Intelligence (AI), in particular to a computing resource optimization method and system for a Convolutional Neural Network (CNN). Background technique [0002] Deep Learning (DL) is a method of simulating the way of thinking of the human brain and dealing with problems. The number of computing neurons in the human brain is on the order of tens of billions, and even a "small" CNN requires huge calculations, and almost all deep learning networks run on CPUs (or CPU clusters), or GPUs (or GPU cluster) hardware platform, the required hardware resources are very huge, resulting in high cost and power consumption, and slow running speed. Many CNNs can only achieve a few frames per second when running on a high-performance chip platform. speed, no real-time processing is possible. [0003] The convolutional neural network includes a convolutional base layer and a f...

AI Technical Summary

Problems solved by technology

[0004] The disadvantages of the above technical solutions are that the computing resources are wasted and the real-time performance of the computing is poor.

Images