Trustworthiness, Privacy, and Security in Federated Learning

In recent years, data privacy security has been widely and highly valued by countries around the world. In the context of European Union’s General Data Protection Regulation (GDPR), the regulatory requirements of laws and regulations are becoming increasingly strict, bringing huge impacts and challenges to enterprises with user’s personal data such as Internet services and financial technology. Up to a point, federal learning ensures data privacy by storing and processing personal data locally. However, due to malicious clients or central servers being able to launch attacks on global models or user privacy data, the security of federated learning is questioned, and introducing blockchain into the federated learning framework is a feasible solution to address these data security issues. In this chapter, the concept of Federated Learning (FL), GDPR, and other similar data protection laws are presented, where the architectures of FL, scale and data partitions in FL, aggregation time schemes, and FL platforms are introduced. In addition, take the Blockchain-empowered Federated Learning (BC-empowered FL) framework as an example, and the commonly used frameworks of BC-empowered FL are introduced, including problem definition, consensus mechanisms, and convergence proofs. Finally, the challenges and directions for future research in the field of Federated Learning are summarized.