电信科学 ›› 2020, Vol. 36 ›› Issue (11): 18-27.doi: 10.11959/j.issn.1000-0801.2020283

• 综述 • 上一篇    下一篇

面向机器学习的隐私保护关键技术研究综述

刘姿杉,程强,吕博   

  1. 中国信息通信研究院,北京 100191
  • 修回日期:2020-08-10 出版日期:2020-11-20 发布日期:2020-12-09
  • 作者简介:刘姿杉(1992- ),女,博士,中国信息通信研究院工程师,主要研究方向为人工智能、宽带接入、数据安全隐私等|程强(1977- ),男,中国信息通信研究院高级工程师,主要研究方向为人工智能、宽带接入、家庭网络等|吕博(1981- ),男,博士,中国信息通信研究院高级工程师,主要研究方向为人工智能、量子计算、高精度时间同步等
  • 基金资助:
    国家重点研发计划基金资助项目(2018YFB1801200)

A survey on key technologies of privacy protection for machine learning

Zishan LIU,Qiang CHENG,Bo LV   

  1. China Academy of Information and Communications Technology,Beijing 100191,China
  • Revised:2020-08-10 Online:2020-11-20 Published:2020-12-09
  • Supported by:
    The National Key Research and Development Program of China(2018YFB1801200)

摘要:

随着信息通信技术的发展,机器学习已经成为多个研究领域与垂直行业必不可少的技术工具。然而,机器学习所需数据中往往包含了大量的个人信息,使其隐私保护面临风险与挑战,受到了越来越多的关注。对现有机器学习下隐私保护法规政策与标准化现状进行梳理,对适用于机器学习的隐私保护技术进行详细介绍与分析。隐私保护算法通常会对数据质量、通信开支与模型表现等造成影响,因此对于隐私保护算法的评估应当进行多维度的综合评估。总结了适用于机器学习应用的隐私保护性能评估指标,并指出隐私保护需要考虑对数据质量、通信开支以及模型准确率等之间的影响。

关键词: 机器学习, 隐私保护, 评估指标

Abstract:

With the development of information and communication technology,large-scale data collection has vastly promoted the application of machine learning in various fields.However,the data involved in machine learning often contains a lot of personal private information,which makes privacy protection face new risks and challenges,and has attracted more and more attention.The current progress of the related laws,regulations and standards to the personal privacy protection and data safety in machine learning were summarized.The existing work on privacy protection for machine learning was presented in detail.Privacy protection algorithms usually have influence on the data quality,model performance and communication cost.Thus,the performance of the privacy protection algorithms should be comprehensively evaluated in multiple dimensions.The performance evaluation metrics for the privacy protection algorithms for machine learning were presented,given with the conclusion that the privacy preservation on machine learning needs to balance the data quality,model convergence rate and communication cost.

Key words: machine learning, privacy protection, performance metric

中图分类号: 

No Suggested Reading articles found!