A Review of Research on Privacy Breaches in Federated Learning

With the rapid development of artificial intelligence technology, federated learning, as a privacy-preserving distributed machine learning paradigm, has attracted widespread attention. FL enables client devices to train models locally and collaboratively generate a global model without directly transmitting data, thereby protecting data privacy to some extent. However, despite its advantages in privacy protection, FL still faces privacy leakage threats, especially in decentralized FL, where client nodes and communication protocols are the primary sources of privacy leakage. This paper systematically investigates privacy leakage issues in FL, analyzes the sources and classifications of privacy leakage in decentralized FL, and explores the applications and effects of existing privacy protection technologies, such as SMPC, HE, and KD. This research fills the gap in privacy protection schemes for decentralized FL and provides references for future research on FL security.