Unlock instant, AI-driven research and patent intelligence for your innovation.

Grouping convolution process optimization method for embedded platform

An embedded and embedded device technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as inability to make full use of embedded device resources, inability to dynamically adjust, and poor scalability, reducing computing power. volume, improve accuracy, and achieve simple results

Inactive Publication Date: 2019-11-29
NORTHWESTERN POLYTECHNICAL UNIV
View PDF5 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, after this method is grouped, the number of channels in each group is fixed, and it cannot be dynamically adjusted according to the resource situation of the embedded device. The scalability is poor, and the resources of the embedded device cannot be fully utilized.

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
  • Grouping convolution process optimization method for embedded platform
  • Grouping convolution process optimization method for embedded platform
  • Grouping convolution process optimization method for embedded platform

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0021] Now in conjunction with embodiment, accompanying drawing, the present invention will be further described:

[0022] The present invention has carried out two aspects of optimization aiming at the traditional packet convolution process, and the specific technical scheme is as follows:

[0023] First, although the traditional grouped convolution can effectively reduce the amount of parameters and reduce computing overhead, if the embedded device still does not have enough resources to meet the computing needs after grouping, the entire network will become very inefficient. In order to further reduce the calculation overhead and retain the characteristics of the data as much as possible, the present invention proposes a gapping method, which combines the data by calculating the mean value, and uses the result as a new input for the entire network. In order to make the result more accurate, the present invention proposes A more uniform merging algorithm is proposed. The ba...

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 relates to a grouping convolution process optimization method for an embedded platform. The channel number of each group is dynamically adjusted according to the resource (calculation and storage capacity) condition of the embedded equipment; the resources of the embedded equipment are fully utilized; when the available resources of the embedded equipment cannot meet the traditionalpacket convolution, a data merging algorithm is adopted, and through the merging of input channels, the calculation amount of the network is further reduced, so that the method can meet the conditionthat the resources of the embedded equipment are limited, and the calculation efficiency is greatly improved; when the available resources of the embedded device meet the requirement of the traditional packet convolution and are remained, a data selection algorithm is adopted, so that the hardware resources of the embedded platform can be used to the maximum extent, and the precision of the network model is further improved.

Description

technical field [0001] The invention relates to the technical field of convolutional neural network optimization on an embedded platform, in particular to a grouped convolution process optimization method for an embedded platform. Background technique [0002] In recent years, thanks to the development of high-performance computer hardware (such as GPU), deep learning, especially convolutional neural network, has been widely used in object detection, image recognition, computer vision and other fields, and has achieved great results. Because the traditional convolutional neural network has high requirements for hardware performance, coupled with the large amount of network parameters and calculations, it cannot be directly deployed on embedded devices with limited hardware resources. Therefore, scholars have proposed a series of Optimization methods, the most representative of which is group convolution. Group convolution first appeared in AlexNet. Due to the limited hardwa...

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): G06N3/04G06N3/063G06N3/08
CPCG06N3/063G06N3/08G06N3/045
Inventor 王继禾黄澍赵佳祥刘君侯雨珊
Owner NORTHWESTERN POLYTECHNICAL UNIV