Neural network convolution method and device

A neural network and convolution technology, applied in the field of neural network convolution, can solve problems such as not supporting NHWC data distribution, and achieve the effect of improved computing performance and excellent computing performance

Active Publication Date: 2020-08-14
SUZHOU LANGCHAO INTELLIGENT TECH CO LTD
View PDF4 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] In view of this, the purpose of the embodiment of the present invention is to propose a method and device for neural network convolution. By using the method of the present invention, the problem of data distribution that does not support NHWC in TVM can be solved, and the calculation of process convolution is realized by using Tensorcore. The performance improvement is more than twice that of TVM's current Direct calculation method, and it can achieve relatively excellent calculation performance in the training and inference process of the AI ​​​​image recognition network model

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Neural network convolution method and device
  • Neural network convolution method and device
  • Neural network convolution method and device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0045] In order to make the object, technical solution and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0046] Based on the above purpose, the first aspect of the embodiments of the present invention proposes an embodiment of a neural network convolution method. figure 1 Shown is a schematic flow chart of the method.

[0047] Such as figure 1 As shown in , the method may include the following steps:

[0048] S1 judges whether the parameters of the image to be input meet the matrix shape calculated by Tensorcore. There are three shapes supported by Tensorcore calculation, and only one of them can be satisfied;

[0049] S2 responds to the parameters of the picture satisfying the matrix shape calculated by Tensorcore, and defines the operator for two-dimensional convolution;

[0050] According to the rule...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention provides a neural network convolution method and device. The method comprises the following steps of judging whether parameters of a to-be-input picture meet a matrix shape calculated byTensorcore or not; in response to the fact that the parameters of the picture meet the matrix shape of Tensorcore calculation, defining an operator of two-dimensional convolution; designing scheduling meeting Tensorcore calculation according to a rule defined by an operator; initializing the accumulator, and loading the input picture into a shared memory; and executing scheduling to obtain a convolution calculation result. According to the scheme of the invention, the problem that NHWC data distribution is not supported in the TVM can be solved; the calculation performance of process convolution realized by using Tensorcore is improved by more than two times than that of the current Direct calculation method of the TVM, and excellent calculation performance can be realized in the trainingand reasoning process of an AI image recognition network model.

Description

technical field [0001] This field relates to the computer field, and more particularly relates to a method and device for neural network convolution. Background technique [0002] With the rapid development of technologies such as the Internet, big data, and cloud computing, the development of artificial intelligence (AI) has advanced by leaps and bounds, and a series of application products have emerged in various fields such as speech recognition, image recognition, intelligent control, and complex computing. It is also widely used in all walks of life, and image processing has always been a research hotspot in the field of artificial intelligence. For example, automatic driving of cars can directly control the driving behavior of cars on the road through the collection and processing of real-time road condition image information; In terms of face recognition, by comparing the data in the information database, the facial features of the face image can be recognized to iden...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/08G06N3/04G06N5/04G06F17/15G06F17/16
CPCG06N3/08G06N5/04G06F17/15G06F17/16G06N3/045
Inventor 王申领
Owner SUZHOU LANGCHAO INTELLIGENT TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap