摘要
近年来,明文数据流通引起的数据泄露及滥用问题日益突出,需要跨域数据融合的机器学习面临新挑战。联邦学习因能在数据本地化的情况下实现联合建模,成为实现数据安全融合的潜在方案。在概述联邦学习理论,分析其金融应用现状的基础上,指出现有联邦学习平台存在安全性不明晰、通用性不足、异构性突出等问题。最后提出健全监管体系、建立通用型技术架构、出台联邦学习行业标准等建议,为推动联邦学习实现金融业数据安全融合提供参考。
In recent years,the problems of data leakage and abuse caused by the circulation of explicit data have become increasingly prominent,and machine learning that requires cross-domain data fusion faces new challenges.Federated learning is a potential solution to achieve data security fusion in that it can realize federated modeling with data localization.Based on an overview of the theory of federated learning and an analysis of the current situation of its financial applications,this article argues that the existing federated learning platforms are faced with the problems of unclear security,insufficient generality,and prominent heterogeneity,etc.It is recommended that we should improve the regulatory system,establish a common technical architecture,and introduce federated learning industry standards to provide references for promoting federal learning to achieve data security integration in the financial industry.
作者
陈琨
李艺
王国赛
时代
杨祖艳
Chen Kun;Li Yi;Wang Guosai;Shi Dai;Yang Zuyan(PBC School of Finance,Tsinghua University,Beijing 100800,China;Huakong Tsingjiao Information Technology(Beijing)Co.,Ltd,Beijing 100129,China)
出处
《征信》
北大核心
2021年第10期29-36,共8页
Credit Reference
关键词
联邦学习
机器学习
数据
隐私保护
金融应用
federated learning
machine learning
data
privacy protection
financial applications