联邦学习应用技术研究综述  

A Review of Federated Learning Application Technologies

作　　者：何泽平 许建[1,2] 戴华 杨庚[1,2] HE Zeping;XU Jian;DAI Hua;YANG Geng(School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;Jiangsu Key Laboratory of Big Data Security and Intelligent Processing,Nanjing 210023,China)

机构地区：[1]南京邮电大学计算机学院,南京210023 [2]江苏省大数据安全与智能处理重点实验室,南京210023

出　　处：《信息网络安全》2024年第12期1831-1844,共14页Netinfo Security

基　　金：国家自然科学基金[62372244]。

摘　　要：人工智能在训练推理等过程中的隐私泄露、推理失真等安全问题,引起了人们的高度关注,甚至涉及意识形态乃至国家战略安全。在此背景下,联邦学习作为一种新兴的机器学习架构,通过保持数据本地性的同时实现模型的联合训练,为多方参与数据分析、处理和共享的应用领域提供了有效的隐私保护能力。从联邦学习的研究动机、技术方法等方面来看,如何利用该技术有效解决典型应用场景下的实际问题是其核心和关键,因此相关应用研究现状的全面综述,对联邦学习的进一步研究与实践都具有参考价值。为此,文章对联邦学习在异常检测、推荐系统以及自然语言处理等典型技术应用中的研究现状进行综合性调研。首先,文章对相关文献按照应用场景角度进行全面的分类梳理,从多领域视角分析了联邦学习架构的研究现状。其次,文章从技术实现的角度,对比分析了各技术领域中不同方案的数据集合、性能特点、评价指标等方面。在此基础上,文章分析总结了联邦学习研究尤其是系统应用面临的关键挑战和发展方向。Security problems,such as privacy leakage and reasoning distortion,arising from training and reasoning in AI have heightened concerns,even involving ideology and national strategic security.As an emerging machine learning architecture,federated learning provides effective privacy protection capabilities for multi-party data analysis,processing,and sharing by achieving global model training while maintaining private data locality.Then,from the perspective of research motivation,technical methods,and other aspects of federated learning,how to apply this technology in typical application scenarios to solve practical problems effectively is its core.Therefore,this article conducted a comprehensive survey on the current research status of application technology of federated learning in typical scenarios,which would be valuable to further research and practice of federated learning.Firstly,a comprehensive classification and sorting of relevant literature were conducted from the perspective of research application scenarios,and the research status in each scenario was analyzed from a multidisciplinary perspective.Secondly,from the perspective of technical implementation,a comparative analysis was conducted on the data sets,performance characteristics,evaluation indicators,and other aspects of different schemes in various application scenarios.Finally,the key challenges and development directions faced by federated learning research,especially system applications,were analyzed and summarized.

关 键 词：联邦学习 异常检测 推荐系统 自然语言处理 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]