FEDERATED MACHINE LEARNING AS A PATHWAY TO PRIVACY-PRESERVING ARTIFICIAL INTELLIGENCE IN SMART DIGITAL ECOSYSTEMS

Abstract

The ever-increasing trend towards the integration of artificial intelligence, edge computing, and hyper-connectivity has seen the emergence of intelligent digital ecosystems encompassing health care, financial sector, municipal administration, IoT, and industry 5.0. However, such ecosystems have to rely on constant large-scale data collection and, hence, the tradition of data storage on one central server becomes more and more challenging in view of privacy concerns, regulatory compliance, and risk appetite of organizations. In order to address this challenge, federated machine learning (FML) has been developed: it makes it possible to train a global machine learning model jointly by multiple participants without collecting any raw data at one central server. This paper provides a systematic narrative review of FML as a basis for privacy-preserving artificial intelligence based on peer-reviewed literature indexed in IEEE Xplore, SpringerLink, ScienceDirect, ACM Digital Library, Wiley Online Library, Taylor & Francis Online, MDPI, Nature Portfolio, and Google Scholar (articles published from 2020 to 2026. Other than summarizing the architectural and algorithmic advancements made in Federated Machine Learning, there are three key contributions that this review makes, setting it apart from previous literature: first, it clearly highlights why and how the scope and synthesis presented in the current review differ from the previous ones; second, it provides an entirely transparent selection process following the PRISMA guidelines, including details on the search strategy used and inclusion/exclusion criteria for selected articles; third, it presents all significant Federated Optimization algorithms in one comparative table instead of presenting them separately in prose. The review highlights the application of Differential Privacy, Secure Multi-Party Computation, and Homomorphic Encryption to provide model updates privacy protection and the use cases of FML in healthcare, smart cities, IoT, finance, and Industry 5.0. Key challenges associated with FML such as non-IID datasets, communication costs, device diversity, adversarial attacks, and fragmented governance are also discussed, along with emerging trends like Explainable AI, blockchain-enabled trust layers, edge intelligence, generative and foundation models, as well as next-generation wireless communication technologies.