个性化联邦学习算法综述OA

Personalized Federated Learning(PFL)is a machine learning framework based on Federated Learning(FL),which enables clients in different domains to protect local data privacy while participating in centralized model training and obtaining personalized models at the same time that conform to local data,but also face significant challenges such as heterogeneity and structural design.We make a comprehensive analysis on the development process,main algorithms,related technologies,shortcomings and future development direction of personalized federated learning.After tracing the origin of personalized federated learning,we discuss the research difficulties and main challenges of personalized federated learning at the present stage,and analyze and compare the main personalized federated learning algorithms from the two dimensions of training algorithms and learning algorithms.Finally,on the basis of discussing the limitations of the current personalized federated learning algorithm,the future development direction of personalized federated learning is prospected,and some research ideas are provided for the corresponding fields.