Method and device for building neural network

A neural network and neural network model technology, applied in neural learning methods, biological neural network models, physical implementation, etc., can solve the problems of difficult neural network training and low efficiency, and achieve easy implementation, simple optimization process, and improved efficiency. Effect

Inactive Publication Date: 2017-10-13
BEIJING TUSEN WEILAI TECH CO LTD
View PDF0 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] The present invention provides a method and device for constructing a neural network to solve the technical problems of high difficulty and low efficiency in training the neural network constructed in the prior art

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
  • Method and device for building neural network
  • Method and device for building neural network
  • Method and device for building neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0029] see figure 1 , is a flowchart of a method for constructing a neural network in an embodiment of the present invention, the method comprising:

[0030] Step 101, constructing an initial neural network, in which a plurality of specific structures preset in the initial neural network are respectively provided with corresponding sparse scaling operators, wherein the sparse scaling operators are used to scale the output of the corresponding specific structures.

[0031] Step 102, using preset training sample data to train the weights of the initial neural network and the sparse scaling operator with a specific structure to obtain an intermediate neural network.

[0032] Step 103 , deleting a specific structure in the intermediate neural network whose sparse scaling operator is zero to obtain a target neural network.

[0033] Preferably, the aforementioned step 101 can be realized through the following steps A1 to A3:

[0034] Step A1, selecting a neural network model.

[...

Embodiment 2

[0083] Based on the same inventive concept as the method for constructing a neural network provided in the first embodiment, the second embodiment of the present invention provides a device for constructing a neural network, the structure of which is as follows Image 6 shown, including:

[0084] The first construction unit 61 is configured to construct an initial neural network, wherein a plurality of specific structures preset in the initial neural network are respectively provided with corresponding sparse scaling operators, wherein the sparse scaling operators are used to perform output of corresponding specific structures Zoom;

[0085] The training unit 62 is configured to use preset training sample data to train the weight of the initial neural network and the sparse scaling operator of a specific structure to obtain an intermediate neural network;

[0086] The second construction unit 63 is configured to delete a specific structure in the intermediate neural network i...

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 discloses a method and device for building a neural network so as to solve a technical problem that a neural network built in the prior art is high in optimization difficulty and low in efficiency. The method comprises the steps of building an initial neural network, wherein a plurality of specific structures preset in the initial neural network are respectively provided with corresponding sparse scaling operators, and the sparse scaling operators are used for scaling the output of the corresponding specific structures; training a weight of the initial neural network and the sparse scaling operators of the specific structures by adopting preset training sample data, and acquiring an intermediate neural network; deleting specific structures with the sparse scaling operators being zero in the intermediate neural network, and acquiring a target neural network. The neural network built by adopting the technical scheme of the invention is simple in optimization, easy to implement and high in training efficiency.

Description

technical field [0001] The invention relates to the field of computers, in particular to a method and device for constructing a neural network. Background technique [0002] In recent years, deep neural networks have achieved great success in many fields, such as computer vision and natural language processing. However, at present, the structure of deep neural networks is mainly designed by designers. To design a deep neural network with compact structure, fast operation speed and good effect, not only requires designers to have strong professional knowledge, but also needs to pass A large number of experiments are used to repeatedly adjust the deep neural network. Therefore, the construction of the existing deep neural network requires high professional skills of designers and the construction efficiency is low. [0003] In order to solve the aforementioned problems in the construction of neural networks in the prior art, some solutions for how to construct deep neural net...

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): G06N3/06G06N3/08
CPCG06N3/063G06N3/08G06N3/061
Inventor 王乃岩黄泽昊
Owner BEIJING TUSEN WEILAI 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