IP Library › Granted Patent US 12,737,678
Granted Patent B2
US 12,737,678 · App. 18/027,371 · Granted Sep 15, 2026

Electronic device and method for federated learning

Inventors: Chen Sun (Beijing, CN); Songtao Wu (Beijing, CN); Tao Cui (Beijing, CN)
Assignee: SONY GROUP CORPORATION
G06N20/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,737,678
App. No.
18/027,371
Granted
Sep 15, 2026
Kind
B2
Abstract

The present disclosure provides an electronic device and a method for federated learning. The electronic device for federated learning at a central processing apparatus comprises a processing circuitry which is configured to: determine a group of distributed nodes for generating a global model parameter among a plurality of distributed nodes, wherein the correlation between local training data of the group of distributed nodes meets a specific correlation requirement; and generate the global model parameter based on local model parameters of the group of distributed nodes, wherein the local model parameters are generated by the group of distributed nodes based on respective local training data thereof.

Claims (57)

1 . An electronic device for federated learning, comprising:

a transmitter; and

a processing circuitry configured to:

determine that a specific distributed node will be used to generate a global model parameter,

wherein a correlation between local training data of the specific distributed node and local training data of other distributed nodes for generating the global model parameter meets a specific correlation requirement; and

control the transmitter to transmit a local model parameter of the specific distributed node to a central processing apparatus,

wherein the local model parameter is generated by the specific distributed node based on its local training data, and

control the transmitter to transmit data collection information of the local training data of the specific distributed node to the central processing apparatus,

wherein the data collection information is information used by the central processing apparatus to determine the correlation between the local training data of the specific distributed node and local training data of other distributed nodes that have been determined to be used to generate the global model the parameter,

wherein the data collection information includes node locations of distributed nodes,

wherein the specific correlation requirement includes the specific distributed node being outside associated exclusive regions of the other distributed nodes that have been determined to be used to generate the global model parameter, and

wherein an associated exclusive region of a distributed node is defined by:

a region in which the distance from each point of the region to the distributed node is less than an exclusive distance, or

a set of a second predetermined number of distributed nodes closest in distance to the distributed node.

2 . The electronic device of claim 1 , wherein the processing circuitry is further configured to:

adjust a parameter for acquiring channel resources according to one or both of a network performance parameter and a local training performance parameter of the specific distributed node.

3 . The electronic device of claim 1 , wherein the processing circuitry is further configured to:

control the transmitter to stop acquiring channel resources for transmitting the local model parameter to the central processing apparatus after receiving an instruction to stop uploading from the central processing apparatus.

4 . The electronic device of claim 1 , wherein the processing circuitry is further configured to:

determine a number of the other distributed nodes that have been determined to be used to generate the global model parameter; and

control the transmitter to stop acquiring channel resources for transmitting the local model parameter to the central processing apparatus after the number is equal to or greater than a first predetermined number.

5 . A method for federated learning performed by an electronic device that comprises a processing circuitry and a transmitter, the method comprising:

determining that a specific distributed node will be used to generate a global model parameter,

wherein a correlation between local training data of the specific distributed node and local training data of other distributed nodes for generating the global model parameter meets a specific correlation requirement; and

controlling the transmitter to transmit a local model parameter of the specific distributed node to a central processing apparatus,

wherein the local model parameter is generated by the specific distributed node based on its local training data; and

controlling the transmitter to transmit data collection information of the local training data of the specific distributed node to the central processing apparatus,

wherein the data collection information is information used by the central processing apparatus to determine the correlation between the local training data of the specific distributed node and local training data of other distributed nodes that have been determined to be used to generate the global model the parameter;

wherein the data collection information includes node locations of distributed nodes,

wherein the specific correlation requirement includes the specific distributed node being outside associated exclusive regions of the other distributed nodes that have been determined to be used to generate the global model parameter, and

wherein an associated exclusive region of a distributed node is defined by:

a region in which the distance from each point of the region to the distributed node is less than an exclusive distance, or

a set of a second predetermined number of distributed nodes closest in distance to the distributed node.

6 . The method of claim 5 , further comprising:

adjusting a parameter for acquiring channel resources according to both of a network performance parameter and a local training performance parameter of the specific distributed node.

7 . The method of claim 5 , further comprising:

controlling the transmitter to stop acquiring channel resources for transmitting the local model parameter to the central processing apparatus after receiving an instruction to stop uploading from the central processing apparatus.

8 . The method of claim 5 , further comprising:

determine a number of the other distributed nodes that have been determined to be used to generate the global model parameter; and

controlling the transmitter to stop acquiring channel resources for transmitting the local model parameter to the central processing apparatus after the number is equal to or greater than a first predetermined number.

9 . A non-transitory computer medium containing instructions for a method for federated learning that is performed by an electronic device that comprises a processing circuitry and a transmitter, the method comprising:

determining that a specific distributed node will be used to generate a global model parameter,

wherein a correlation between local training data of the specific distributed node and local training data of other distributed nodes for generating the global model parameter meets a specific correlation requirement; and

controlling the transmitter to transmit a local model parameter of the specific distributed node to a central processing apparatus,

wherein the local model parameter is generated by the specific distributed node based on its local training data; and

controlling the transmitter to transmit data collection information of the local training data of the specific distributed node to the central processing apparatus,

wherein the data collection information is information used by the central processing apparatus to determine the correlation between the local training data of the specific distributed node and local training data of other distributed nodes that have been determined to be used to generate the global model the parameter;

wherein the data collection information includes node locations of distributed nodes,

wherein the specific correlation requirement includes the specific distributed node being outside associated exclusive regions of the other distributed nodes that have been determined to be used to generate the global model parameter, and

wherein an associated exclusive region of a distributed node is defined by:

a region in which the distance from each point of the region to the distributed node is less than an exclusive distance, or

a set of a second predetermined number of distributed nodes closest in distance to the distributed node.

10 . The electronic device of claim 1 , wherein the processing circuitry is further configured to determine that the specific distributed node will be used to generate the global model parameter based on an instruction to upload the local model parameter from the central processing apparatus.

11 . The method of claim 5 , further comprising:

determining that the specific distributed node will be used to generate the global model parameter based on an instruction to upload the local model parameter from the central processing apparatus.

12 . The non-transitory computer medium of claim 9 , wherein the method further comprises:

determining that the specific distributed node will be used to generate the global model parameter based on an instruction to upload the local model parameter from the central processing apparatus.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Mar 21, 2023
From: SUN, CHEN; WU, SONGTAO; CUI, TAO
To: SONY GROUP CORPORATION
Reel/Frame 063040/0943 →
Priority Claims (1)
CN 202011173054.X · Oct 28, 2020 · national
Continuity (1)
Related Publication 20230385688A1 · Nov 30, 2023
References Cited (16)
US 10003502B1 · Bernardo De Oliveira · 2018 [cited by applicant]
US 20180373956A1 · Yamato · 2018 [cited by applicant]
US 20180373988A1 · Dhanyamraju · 2018 [cited by examiner]
US 20200219010A1 · Jobling · 2020 [cited by examiner]
US 20250209383A1 · Hu · 2025 [cited by examiner]
CN 110167176A · 2019 [cited by examiner]
CN 111090631A · 2020 [cited by examiner]
CN 111325351A · 2020 [cited by examiner]
CN 111369042A · 2020 [cited by examiner]
CN 111754000A · 2020 [cited by examiner]
CN 111784002A · 2020 [cited by examiner]
CN 215117297U · 2021 [cited by examiner]
JP 2017111501A · 2017 [cited by applicant]
JP 2022524662A · 2022 [cited by examiner]
WO WO2019112667A1 · 2019 [cited by examiner]
International Search Report and Written Opinion mailed on Jan. 26, 2022, received for PCT Application PCT/CN2021/126875, filed on Oct. 28, 2021, 9 pages including English Translation. [cited by applicant]