«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年10期

页码:

1-7

栏目:

综述

出版日期:

2024-10-10

文章信息/Info

Title:

Survey and Analysis of Privacy-preserving Based on Horizontal Federated Learning

文章编号:

1673-629X(2024)10-0001-07

作者:

王斌1; 2; 尹晓雅1; 张磊1*

1. 佳木斯大学 信息电子技术学院 黑龙江省自主智能与信息处理重点实验室,黑龙江 佳木斯 154007;2. 佳木斯大学 佳木斯大学科技处,黑龙江 佳木斯 154007

Author(s):

WANG Bin1; 2; YIN Xiao-ya1; ZHANG Lei1*

1. Key Laboratory of Autonomous Intelligence and Information Processing of Heilongjiang Province,School of Information & Electronic Technology,Jiamusi University,Jiamusi 154007,China;2. Science and Technology Department of Jiamusi University,Jiamusi University,Jiamusi 154007,China

关键词:

隐私保护技术; 隐私安全; 联邦学习; 同态加密; 差分隐私

Keywords:

privacy preserving technology; privacy and security; federated learning; homomorphic encryption; differential privacy

分类号:

TP391.1

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0194

摘要:

近年来,随着人工智能的快速发展,机器学习成为处理数据的重要手段。 传统的机器学习在处理数据时需要上传到中心服务器,无法保证用户数据隐私。 但联邦学习的分布式特性保证了数据安全和隐私,同时也解决了数据孤岛问题,提高了不同应用环境下数据的可用性。 然而,越来越多的研究表明,基于联邦学习的方法仍然遭受不同的隐私威胁,因此如何保护联邦学习场景下的用户数据隐私成为一个挑战。 为应对联邦学习的隐私问题,对现有的隐私保护手段进行了调查分析。 首先,介绍了联邦学习的定义、发展和分类;其次,介绍联邦学习的系统架构并分析了所遭受的隐私威胁手段;再次,根据联邦学习训练过程的生命周期对隐私保护技术进行分类和整理,同时比较了各技术的优缺点;最后,指出了目前联邦学习所面临的挑战,以及未来的发展趋势。

Abstract:

In recent years,with the rapid development of artificial intelligence,machine learning has become an important approach for data processing. Traditional machine learning methods require data to be uploaded to central servers,which raises concerns about data privacy. However,the distributed nature of federated learning ensures data security and privacy,while also addressing the issue of data silos and improving data availability in different application environments. Nevertheless,increasing research indicates that federated learning methods still face various privacy threats,making it a challenge to protect user data privacy in federated learning scenarios. To address the privacy issues in federated learning, we investigated and analyzed existing privacy protection mechanisms. Firstly, we introduced the definition,development,and classification of federated learning. Secondly,presented the system architecture of federated learning and analyzed the privacy threats it faces. Furthermore, classified and organized privacy protection techniques based on the lifecycle of the federated learning training process, while comparing the advantages and disadvantages of each technique. Finally,highlighted the current challenges faced by federated learning and discussed future development trends.

常用功能

更新日期/Last Update: 2024-10-10