Summary - F.748.35 (06/2024) - Requirement and framework of trustworthy federated machine learning based service

Federated machine learning (FML) is an emerging distributed machine learning paradigm that enables collaborative model training, learning, utilization and construction from a large number of distributed datasets on the basis of ensuring data security and legal compliance. In FML, computing takes place where the data are, and although the data are available, neither the data computing nor the data are visible. There are some challenges for FML-based services in aspects of trust as they perform in distributed or decentralized environments. All the challenges are often brought about by a lack of trust in the multiple participants of FML-based services, usually in the processes of model training and utilization, such as data indexing, data computing, parameter exchanging, etc.
Specific functional components are needed to enhance the trustworthiness of FML-based services, such as enhancing dataset indexing, data computing, parameter exchange and model utilization. The distributed ledger technology (DLT) system is one type of trustworthy shared data system that can be used to also store FML-based service data. The FML-based service can take advantage of the convergence between FML and those components, especially to help address trust-related challenges for FMLbased services.
Recommendation ITU-T�F.748.35 provides a trustworthy FML-based service, and specifies its concept, general characteristics and requirements, and reference framework.