南京邮电大学学报（自然科学版）2025，Vol.45Issue(3)：99-108,10.DOI:10.14132/j.cnki.1673-5439.2025.03.011
面向开放世界的联邦学习综述:挑战、技术与应用
A survey on federated learning in the open world:challenges,technologies and applications
摘要
Abstract
With the increasing awareness of data privacy protection and the continuous changes in practi-cal application environments,federated learning has attracted significant attention as a privacy-preserving machine learning method.Real-world scenarios are often open and dynamic,making federated learning one of the research hotspots.This paper comprehensively reviews relevant issues faced by feder-ated learning in open environments,systematically classifies existing methods,and explores their poten-tial applications in network traffic classification.Firstly,regarding the issue of non-independently and identically distributed(Non-IID)data,solutions in traditional scenarios are reviewed,and dynamic fed-erated learning scenarios are introduced.Secondly,given the scarcity of data labels,relevant algorithms for federated semi-supervised learning are summarized.Finally,the methods for processing unknown classes in the federated environment are discussed,and future research directions and application pros-pects in network traffic classification are anticipated.关键词
联邦学习/非独立同分布/联邦半监督学习/联邦开放集识别Key words
federated learning/non-independent and identically distributed(Non-IID)/federated semi supervised learning/federated open set recognition分类
信息技术与安全科学引用本文复制引用
陆浩天,董育宁,卢官明..面向开放世界的联邦学习综述:挑战、技术与应用[J].南京邮电大学学报（自然科学版）,2025,45(3):99-108,10.基金项目
国家自然科学基金(61271233)和江苏省研究生科研创新计划(KYCX23_1031)资助项目 （61271233）
