Joint training method based on multi-party 3D printing database

A 3D printing and database technology, applied in electrical digital data processing, instruments, design optimization/simulation, etc., can solve problems such as application and development limitations, and achieve the effects of easy convergence, reduced transmission volume, reduced demand and input costs

Active Publication Date: 2022-04-22
CHENGDU AIRCRAFT INDUSTRY GROUP
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Industrial design, architecture, automobile, aerospace, dentistry, education fields, etc. are all applied, but its application and development are still limited by factors

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
  • Joint training method based on multi-party 3D printing database
  • Joint training method based on multi-party 3D printing database
  • Joint training method based on multi-party 3D printing database

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0034] A method for joint training based on multi-party 3D printing databases, including multiple training members and servers, the model of the training members is wi , the data for each training member is Xi , , labeled as y i ; the server's model is W , and the server's model W Model with training members wi The network structure is the same; including the following steps:

[0035] Step S100: at j During training iterations, the training members i read Xi one of batch The data bi , and model wi The forward propagation of to get the predicted label , and then according to the actual label y i , the calculated model wi The loss function, and then use the backpropagation algorithm to get the gradient matrix G i ;

[0036] Step S200: train members i pair gradient matrix G i The elements in are sorted from large to small according to the absolute value, and the first m elements are selected to obtain the corresponding sparse matrix , the filling elem...

Embodiment 2

[0040] This embodiment is optimized on the basis of Embodiment 1. Before the iterative training, the model is initialized: the server to the model W Initialize and distribute the initialization results to all training members, and the model W i Perform initialization to determine the gradient upload scale coefficient α, attenuation coefficient ρ, and learning rate γ.

[0041] Further, in the step S200, the model is obtained statistically W i The total number of elements in is M , calculate the number of gradient elements that need to be uploaded this time .

[0042] Other parts of this embodiment are the same as those of Embodiment 1, so details are not repeated here.

Embodiment 3

[0044] This embodiment is optimized on the basis of Embodiment 1 or 2, and the correlation calculation in the step S300 is as follows:

[0045]

[0046] in:

[0047]

[0048] in: D KL for KL Divergence,

[0049] P Indicates the quality of each training member's own data,

[0050] Q Indicates the data quality of all samples of the server.

[0051] Other parts of this embodiment are the same as those of Embodiment 1 or 2 above, so details are not repeated here.

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 joint training method based on a multi-party 3D printing database, and the method comprises the steps: obtaining a gradient matrix Gi of a training member i in a jth training iteration process; the training member i sorts elements in the gradient matrix Gi according to absolute values, first m elements are selected to obtain a sparse matrix, and a filling element is 0; calculating the relevancy between the data quality Pi of each training member and the data quality Q corresponding to all samples of the server, carrying out sorting based on the relevancy, and obtaining the training members participating in the current iteration; and finally, updating the model of the server by using the sparse matrix of the training members participating in the current iteration, and correspondingly updating the model Wi of the training members. According to the method, joint training among multiple databases is realized, the training members of the current round of iteration are determined based on the relevancy, the data transmission amount in the joint training process is reduced, the requirement for the data transmission bandwidth and the input cost are reduced, and the method has good practicability.

Description

technical field [0001] The invention belongs to the technical field of joint processing of print data, and in particular relates to a joint training method based on multi-party 3D printing databases. Background technique [0002] In the 1980s, 3D printing technology was born. 3D printing is not limited to the traditional "removal" processing method, but 3D printing is a bottom-up manufacturing method, also known as additive manufacturing technology, which realizes The establishment of mathematical models. 3D printing technology has been widely concerned by people since its birth, so it has developed rapidly. In recent decades, 3D printing technology has become the focus of attention. It is applied in industrial design, architecture, automobile, aerospace, dentistry, education fields, etc., but its application and development are still limited by factors. [0003] In the implementation process of 3D printing, due to too many 3D printing related parameters, it is impossible...

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): G06F30/27G06N20/20G06F113/10
CPCG06F30/27G06N20/20G06F2113/10
Inventor 荣鹏高鹏高川云杜娟
Owner CHENGDU AIRCRAFT INDUSTRY GROUP
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