Connecting the world and beyond

Federated Learning & Distributed AI

8.0.png Dedicated to solving the "data silo" problem between different institutions and the privacy need to keep data locally. It formulates a series of security collaboration standards regulating Federated Machine Learning (FML) architectures, node contribution quantification, heterogeneous resource control, and distributed intelligent model communication components.

8.1 Federated Learning Frameworks

Establishes federated learning and federated foundation model distributed training architecture systems covering basic trusted environments, cloud-edge-end collaborative load distribution, and fully decentralized environments.

ITU-T F.748.35: Requirement and framework of trustworthy federated machine learning based service (2024)
Proposes a foundational capability framework for constructing trustworthy federated machine learning services, ensuring that underlying sensitive datasets are not leaked during multi-party joint training.

The figure below shows the use case that FML-based autonomous driving service consists of three entities: an autonomous driving terminal (FML participant), a trustworthy shared data storage (DB) system, and an autonomous driving system (FML coordinator).

F.748.35 Trustworthy FML based service in autonomous vehicle environment

ITU-T F.748.62: Device-edge-cloud collaborative federated machine learning framework (2025)
Formulates architecture specifications for multi-tier joint federated machine learning collaborative training built jointly by end devices, edge servers, and cloud central servers.

ITU-T F.FLC-Heter-Arch: Functional architecture of federated learning controller based on heterogeneous resources (2025)
Targeting multi-node environments with highly heterogeneous computing power, networks, and datasets, standardizes the task decomposition and collaborative scheduling architecture for federated learning central controllers.

ITU-T F.FFM: Requirements and Framework for Federated Foundation Model (2025)
Provides flexible federated foundation model architectures and task execution paradigm specifications for distributed joint fine-tuning and training of large pre-trained models with extremely high privacy protection requirements.

8.2 Federated Learning Evaluation

Primarily targets the process contributions of multi-party collaboration in federated learning, providing trusted quantitative measurement and evaluation specifications supported by decentralized ledgers.

ITU-T F.748.24: Trusted contribution evaluation framework on federated machine learning services (2024)
Based on distributed ledger technology, provides a trusted quantitative contribution tracking system and evaluation service workflow for data samples and computing power inputs from various coordinating nodes in federated learning.

In order to build a safe public traffic system, public security departments, traffic management departments and automobile enterprises gather data to train image recognition and traffic prediction models by FML services. Public security departments, traffic management departments and automobile enterprises act as FML training participants, and regulatory departments act as FML training coordinators. A CES manages FML training tasks and acts as a bridge between coordinators of and participants in FML training to exchange training data. In addition, a CES stores the exchanged training data in a DLT system. A CES also manages itself and stores data in DLT systems. DLT systems guarantee the reliability, integrity, consistency, availability and immutability of data. As a result, a CES, supported by DLT system(s), provides a trusted contribution degree

evaluation service for each participant. See Figure below.

F.748.24 Using a CES to build a safe public traffic system 

ITU-T H.FML-AC: Assessment criteria for federated machine learning platforms (2024)
Formulates comprehensive evaluation and testing specifications for federated machine learning platforms across dimensions like account management, node control, and encryption/decryption performance.

ITU-T H.FMLS-QMEM: Quality Metrics and Evaluation Methods for Federated Machine Learning Services (2025)
Defines end-to-end actual operation Quality of Service (QoS) evaluation metrics and verifiable verification methods for federated learning services.

8.3 Distributed AI Architectures

Builds standardized internal reference architectures and protocol exchange interfaces for distributed model computing, feature semantic-level reasoning, and heterogeneous communication systems in ubiquitous computing networks.

ITU-T F.748.13: Technical framework for a shared machine learning system (2021)
Defines the role division and control logic framework when multiple parties rely on shared network computing resources to conduct distributed machine learning collaborative training.

ITU-T F.748.21: Requirements and framework for feature-based distributed intelligent systems (2022)
Establishes the basic functional requirements for distributed feature intelligent analysis systems where feature descriptions are extracted at the front end and aggregated to the central side for post-processing like classification.

ITU-T F.748.22: Functional architecture for feature-based distributed intelligent systems (2023)

Further clearly defines the internal front-end, algorithm, and control functional module groups of Feature-based Distributed Intelligent Systems (FDIS) and various reference point protocol interface specifications.

F.748.22 A functional architecture for feature-based distributed intelligent systems

ITU-T F.748.37: Requirements and functional framework of joint semantic query system of unstructured data across clusters (2024)
Plans the convergence architecture requirements for extracting semantic features and jointly retrieving interactive multi-modal unstructured data (like images and text) across different heterogeneous data clusters.

ITU-T H.626.8: Protocols for feature-based distributed intelligent systems (2025)
Standardizes the underlying encapsulation protocol mechanisms for HTTP-based device signalling and information query interactive communication for feature-based distributed intelligent analysis systems.

ITU-T F.FUCL: Requirements and Framework of Cross-Platform Unified Communication Libraries for Distributed Multimedia AI Systems (2025)

Regulates unified underlying high-performance communication library middleware interface standards across heterogeneous hardware accelerators for distributed training networks.