Summary

Data privacy and information security pose significant challenges to the big data and artificial intelligence (AI) community as these communities are increasingly under pressure to adhere to regulatory requirements. Many routine operations in big data systems and applications, such as merging user data from various sources to build a machine learning model, are considered to be illegal under current regulatory frameworks. The purpose of federated machine learning (FML) is to provide a viable solution that empowers machine learning applications to utilize data in a distributed manner. In an FML framework, the data owners do not exchange raw data directly and do not allow any party to infer the private information of other parties. In order to facilitate the construction and use of federated machine learning models (FMLMs) and improve the quality of FML service, Recommendation ITU-T M.3387 specifies the management requirements for federated machine learning systems (FMLSs), including the functional architecture of FMLSs, as well as the requirements of the basic management domain, model management domain, and data management domain. This Recommendation is applicable to the architecture design, research, and development of FMLSs.