Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
FEDERATED LEARNING METHOD AND COMMUNICATION APPARATUS
Document Type and Number:
WIPO Patent Application WO/2024/103291
Kind Code:
A1
Abstract:
Provided in the present application are a federated learning method and a communication apparatus. The federated learning method comprises: an access network device receiving cumulative dual variable change parameters of a task b, which are sent by a plurality of terminal devices; the access network device determining a dual variable change parameter mean value of the task b on the basis of the cumulative dual variable change parameters of the task b, which are sent by the plurality of terminal devices; the access network device determining updated model parameters of the task b on the basis of the dual variable change parameter mean value of the task b; the access network device sending the updated model parameters of the task b to a server; the access network device receiving updated multi-task constraint parameters, which are sent by the server; and the access network device sending the updated model parameters of the task b and the updated multi-task constraint parameters to the plurality of terminal devices. On the basis of the method of the present application, multi-task federated learning and hierarchical federated learning can be combined, such that the calculation difficulty of multi-task federated learning is reduced while the number of tasks is reduced.

Inventors:
YI XINFU (CN)
LI RONGPENG (CN)
WANG FEI (CN)
PENG CHENGHUI (CN)
Application Number:
PCT/CN2022/132258
Publication Date:
May 23, 2024
Filing Date:
November 16, 2022
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
HUAWEI TECH CO LTD (CN)
International Classes:
H04L27/00
Foreign References:
CN114020469A2022-02-08
CN112766493A2021-05-07
CN113900796A2022-01-07
CN115293358A2022-11-04
CN113194489A2021-07-30
US20210089878A12021-03-25
Other References:
VIRGINIA SMITH: "Federated multi-task learning", 31ST CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS (NIPS 2017), 1 January 2017 (2017-01-01), pages 1 - 11, XP093169867
Attorney, Agent or Firm:
SCIHEAD IP LAW FIRM (CN)
Download PDF: