ITU-T Work Programme

[2022-2024] : [SG16] : [Q5/16]

[Declared patent(s)]

Work item:

F.DEC-CFL

Subject/title:

Technical framework and requirements for device-edge-cloud collaborative federated learning

Status:

[Carried to next study period]

Approval process:

AAP

Type of work item:

Recommendation

Version:

New

Equivalent number:

Timing:

Liaison:

ITU-T SG13

Supporting members:

China Mobile, China Telecom, BJTU China, ZTE China

Summary:

Federated learning (FL) is an emerging distributed framework that enables multiple participants to collaboratively training a shared machine learning (ML) model with the assistance of a server via exchanging model parameters rather than directly sharing decentralized datasets, which can break data islands and meanwhile bring the advantage of data privacy protection and legal compliance. In real-world applications, it is impossible to coordinate large-scale end devices with the remote cloud server because of high load, poor system scalability and single point of failure. Therefore, it is necessary to design device-edge-cloud collaborative FL framework, which introduces edge servers as the intermediate coordinators. This collaborative framework usually consists of one cloud server, multiple edge servers and a multitude of end devices. Each edge device can access a number of end device according to its communication and computation capabilities and the cloud server can access multiple edge servers. As the number of edge devices increases, the accessed number of end devices in this system can enlarge. As a result, this framework has the advantages of high training accuracy, balanced resource utilization and large-scale devices access. This draft new Recommendation specifies the framework and requirements for device-edge-cloud collaborative FL, and specifies its concept, procedure, requirements and use cases.

Comment:

Reference(s):

[SG16-TD263/WP3 (2024-04)

]