Federation learning model training method and device and federation learning system

A technology of learning models and training methods, which is applied in the information field, can solve problems such as gradient offset, achieve the effect of reducing data movement, good effect, and satisfying privacy protection and data security

Active Publication Date: 2021-01-15
ZHEJIANG LAB
View PDF6 Cites 30 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the embodiments of the present invention is to provide a federated learning model training method, device, and federated learning system to solve the gradient offset problem existing in existing solutions

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
  • Federation learning model training method and device and federation learning system
  • Federation learning model training method and device and federation learning system
  • Federation learning model training method and device and federation learning system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0045] figure 1 A schematic diagram of the composition and structure of a federated learning system according to an embodiment of the present invention is shown. Such as figure 1 As shown, the federated learning system mainly consists of a cloud federated learning subsystem, an edge computing server and consists of end devices, among which, Each edge computing server is connected to the cloud federated learning subsystem, and each edge computing server is connected to one or more end devices. , one or more end devices 10 form a regional federated learning group together with an edge computing server 20, and the edge computing server 20 is responsible for multiple rounds of gradient updates of the end devices 10 in the area; one or more edge computing servers 20 are connected to the cloud federation The learning subsystem 30, the cloud federated learning subsystem 30 is responsible for updating the gradient of the edge computing server 20.

Embodiment 2

[0047] figure 2 A schematic diagram showing the composition and structure of end devices of a federated learning system according to an embodiment of the present invention. Such as figure 2 As shown, the terminal device 10 includes a controller, and the controller 110 is connected to at least one 3D stack memory 120, at least one non-volatile memory 140 and a first connection module 130; the controller 110 is responsible for the resources of the entire terminal device 10 Management and local training of federated machine learning; the 3D stack memory 120 is used as the main memory, and it also adopts the near data processing method to have reasoning ability; the non-volatile memory 140 provides the storage function, and also has the function of implementing in-situ calculation; the first connection The module 130 establishes a connection with the edge computing server 20 and receives or transmits information.

Embodiment 3

[0049] image 3 A schematic diagram showing the composition and structure of an edge computing server of a federated learning system according to an embodiment of the present invention. Such as image 3 As shown, the edge server 20 is mainly composed of a processing module 220, a second connection module 210, a communication module 230, a configuration module 240 and a first storage module 250, and the processing module 220 implements the system settings of the edge computing server 20 through the configuration module 240; The processing module 220 establishes a connection with the first connection module 130 of the end device 10 through the second connection module 210; the processing module 220 establishes a network communication connection with the cloud federated learning subsystem 30 through the communication module 230; the first storage module 250 is responsible for storing system software And related data to support the processing module 220 to implement regional fede...

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 federated learning model training method and device and a federated learning system. An edge computing server and end equipment receive global machine learning model information formed by a cloud federated learning subsystem; one edge computing server and more than one end device form an area by utilizing the advantage of network locality, and the end device completes model local training by depending on local data and adopting a truncation mode. The edge computing server is responsible for multiple rounds of updating of end equipment in a jurisdiction area and sending updated model information to the cloud federated learning subsystem; the edge computing servers also complete model local training in a truncation mode, and the cloud federated learning subsystem isresponsible for gradient updating of the edge computing servers; and when the training reaches a convergence period, performing truncation node compensation on the inner-end equipment of the region under the jurisdiction of the edge computing server and a plurality of edge computing servers in charge of the cloud federated learning subsystem to form global machine learning model information.

Description

technical field [0001] The present invention relates to the field of information technology, in particular to a federated learning model training method, device and federated learning system. Background technique [0002] The device-edge-cloud hybrid architecture is a new architecture that supports collaborative computing. Among them, "end" refers to terminal devices (or simply "end devices"), for example, cameras installed in every corner of the city, and mobile phones used by people; "side" refers to edge computing, for example, edge servers installed and deployed near the city; "Cloud" refers to cloud computing, eg, a large data center. Deploying federated learning applications using a device-edge-cloud hybrid architecture can not only protect data privacy, but also reduce data movement. The outstanding problem is that when the data samples of the training participants are unevenly distributed, or the data is skewed, that is, in the case of non-independent and identical...

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): G06N20/20G06F21/60
CPCG06F21/602G06N20/20
Inventor 曾令仿银燕龙何水兵毛旷杨弢任祖杰陈刚
Owner ZHEJIANG LAB
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