Abstract
In recent years, privacy and security concerns in machine learning havepromoted trusted federated learning to the forefront of research. Differentialprivacy has emerged as the de facto standard for privacy protection infederated learning due to its rigorous mathematical foundation and provableguarantee. Despite extensive research on algorithms that incorporatedifferential privacy within federated learning, there remains an evidentdeficiency in systematic reviews that categorize and synthesize these studies. Our work presents a systematic overview of the differentially privatefederated learning. Existing taxonomies have not adequately considered objectsand level of privacy protection provided by various differential privacy modelsin federated learning. To rectify this gap, we propose a new taxonomy ofdifferentially private federated learning based on definition and guarantee ofvarious differential privacy models and federated scenarios. Our classificationallows for a clear delineation of the protected objects across variousdifferential privacy models and their respective neighborhood levels withinfederated learning environments. Furthermore, we explore the applications ofdifferential privacy in federated learning scenarios. Our work provide valuableinsights into privacy-preserving federated learning and suggest practicaldirections for future research.