Differential Privacy and Federated Learning Technologies in Privacy Protection for Medical Big Data

With the rapid development of big data technology, the value of data in various industries has become increasingly prominent, but the problem of privacy leakage has also received more and more attention. How to protect user privacy while mining the value of data has become an important problem that urgently needs to be solved in the field of big data analysis. This paper reviews and summarizes research related to privacy-preserving big data analysis from three levels: the algorithm level, the framework level, and the application level. First, the application and optimization methods of differential privacy (DP) in deep learning model training are introduced, including key technologies such as the differentially private stochastic gradient descent (DP-SGD) framework, gradient clipping, and privacy budget optimization. Then, the problem of improving model efficiency under the federated learning (FL) framework is discussed, with a focus on communication efficiency optimization and personalized FL methods under non-independent and identically distributed (non-IID) data. Finally, taking medical big data as a typical scenario, the research status and challenges of privacy protection systems are discussed. At the end of the paper, the existing problems in current research and future development directions are also preliminarily prospected.