Introduction to Federated Learning and Threat Landscape

Federated Learning (FL), a collaborative machine learning paradigm that allows model training across multiple devices without centralized data aggregation, was developed in response to the exponential growth of decentralized data across mobile devices, edge sensors, and distributed systems. By limiting sharing to model updates and preserving sensitive user data locally, FL overcomes important privacy, legal, and bandwidth restrictions. However, new and advanced security threats are introduced by this distributed and privacy-aware architecture. These can compromise the reliability and effectiveness of federated models and include poisoning attacks, inference leakage, free-riding, and Byzantine behavior. This chapter provides overview of FL with real-world applications. The chapter is designed to provide readers with a solid knowledge base on the opportunities and challenges in the construction of reliable, scalable, and elastic FL systems, eliminating the gap between the Federated Model design and thoughtfulness.