Federated Learning in Agents Based Cyber-Physical Systems

The pervasiveness of connected wearable devices, combined with recent advances in Artificial Intelligence (AI), which is capable of extracting significant insights from vast amounts of collected data, has driven the development of cyber-physical systems composed of human users and intelligent agents. Multi-agent systems (MAS) represent a powerful paradigm to develop such kind of distributed systems. They can autonomously learn from data and assist users to manage personal information, supporting individuals in their daily activities. In the smart-health context, sharing personal health data with physicians can create information flows that enhance the processes of diagnosis, monitoring, and treatment for a growing number of conditions, particularly in frail patients. However, with increasing emphasis on data privacy and security, there is heightened awareness of the risks associated with collecting data from all users for machine learning tasks. Federated Learning (FL) has emerged as a solution, offering a decentralized framework where a shared prediction model can be built while ensuring that owners’ data remains on their own devices. In FL, the data residing on various devices may exhibit imbalances and non-i.i.d. (independent and identically distributed) characteristics. In this chapter, we introduce the main concepts of the FL paradigm and the relevance of MAS as an enabling technology to exploit FL in decentralized environments such as edge devices or IoT networks.