Description
A Federated Learning (FL) member, also referred to as an FL participant, is a functional entity that plays a defined role within a federated learning process in the 3GPP AI/ML ecosystem. Federated learning is a distributed machine learning paradigm in which multiple parties collaboratively train a shared model without exchanging their raw training data. Instead, each participant trains a local model on its own data and shares only model updates (e.g., gradients or weights) with an aggregator. 3GPP introduced the FL member concept in the Rel-19 study on AI/ML (Artificial Intelligence / Machine Learning) support in the 5G core network, documented in TR 23.700 and TS 23.482.
The FL member concept distinguishes two principal roles. An FL client is the entity that performs local ML model training on locally available data and reports the resulting model updates to the FL server. An FL server is the entity that performs aggregation and collaboration functions for the FL process, combining updates received from multiple FL clients into an improved global model and distributing the updated model back to the clients. A single FL member may in principle take on either role depending on the deployment and the phase of the FL procedure.
In the 3GPP architecture, the FL member is assumed to be realized either as a functionality hosted at the VAL UE (Vertical Application Layer User Equipment) or at the network/server side, for example within an AIML enablement server or a VAL server. This placement flexibility reflects the fact that federated learning can be applied both to data generated at the device (e.g., user behavior, sensor readings) and to data held in the network (e.g., analytics from NWDAF or other core network functions). The FL member therefore sits at the boundary between the vertical application layer and the AI/ML enablement infrastructure.
The FL member interacts with other FL-related entities through the FL process, which includes procedures for FL session establishment, FL client selection, model distribution, local training, model update reporting, and aggregation. Security and privacy considerations are central: because raw data never leaves the FL member, federated learning provides a privacy-preserving alternative to centralized training. 3GPP specifications also consider aspects such as authorization of FL members, discovery of FL clients and servers, and the interaction between FL and the 5G core network's AI/ML capabilities.
The concept matters because it provides a standardized vocabulary and architectural anchor for describing who does what in a federated learning deployment spanning UEs, vertical servers, and network functions. Without such a definition, interoperability between different vendors' FL implementations and between the network and vertical domains would be difficult to specify and test.
Purpose & Motivation
Federated learning exists to enable collaborative machine learning across distributed data sources without centralizing sensitive data. Traditional centralized ML requires moving raw data to a central training server, which raises privacy, regulatory, and bandwidth concerns, particularly when data originates from user devices or from multiple administrative domains. The FL member concept was created in 3GPP Rel-19 to give a precise, standardized name to the entities that participate in this distributed training process, so that the 5G system can describe how UEs, vertical application servers, and network-side AI/ML functions cooperate.
Before the introduction of this term, 3GPP AI/ML work focused largely on network-side analytics and model training within the core network (e.g., NWDAF-based analytics). As verticals began to demand on-device learning and privacy-preserving collaboration, a common vocabulary was needed to describe the roles of data owners and aggregators across the UE, VAL, and network domains. The FL member definition addresses this by explicitly allowing the role to be hosted at the VAL UE or at the network/server side, thereby accommodating both device-centric and network-centric FL deployments.
The definition also supports the specification of FL procedures such as client selection, model distribution, and aggregation, which require clear identification of which entity performs training and which performs aggregation. By anchoring these procedures to a well-defined entity, 3GPP enables interoperable, secure, and privacy-respecting federated learning across multi-vendor 5G networks and vertical applications.
Release Timeline
Detected Changes Across Releases
from 3GPP Change RequestsSpecific changes extracted from the „Change history“ tables of 3GPP specifications (9 CRs across 2 releases). Complements the general historical overview above with the evidence-based evolution of this function.
Defining Specifications
3GPP specifications that define or reference FL member, with the latest known release. Sourced from the 3GPP document catalog — see methodology.
| Specification | Title | Release |
|---|---|---|
| TS 23.482 vk30 | AI/ML Enablement for Verticals | Rel-20 read |
| TS 23.700 vk10 | AI/ML Application Layer Support Phase 2 | Rel-20 read |