Article(id=1245407858494390590, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156262727438951343, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2308502, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1698681600000, receivedDateStr=2023-10-31, revisedDate=1721923200000, revisedDateStr=2024-07-26, acceptedDate=null, acceptedDateStr=null, onlineDate=1774857972023, onlineDateStr=2026-03-30, pubDate=1741363200000, pubDateStr=2025-03-08, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1774857972023, onlineIssueDateStr=2026-03-30, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1774857972023, creator=13701087609, updateTime=1774857972023, updator=13701087609, issue=Issue{id=1156262727438951343, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='7', pageStart='2193', pageEnd='3077', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=0, createTime=1753604116544, creator=13701087609, updateTime=1753771263994, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1156963794699248405, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156262727438951343, language=EN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1156963794699248406, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1156262727438951343, language=CN, specialIssueTitle=, coverIllustrator=, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=2849, endPage=2855, ext={EN=ArticleExt(id=1245407859123536199, articleId=1245407858494390590, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=Adaptive Privacy Protection Scheme for Federated Learning Based on Differential Privacy, columnId=1156262729162810294, journalTitle=Science Technology and Engineering, columnName=Papers·Automation and Computational Technology, runingTitle=null, highlight=null, articleAbstract=

With the deepening research on federated learning, it has been observed that the privacy protection strategies employed within federated learning fall short of fully guaranteeing the security and confidentiality of user data. Moreover, the training process in federated learning encounters challenges regarding model convergence. In response to these aforementioned issues, an innovative solution termed adaptive differential privacy (DP-AdaMod) was proposed. Primarily, the model training process was fine-tuned by incorporating an adaptive learning rate algorithm to mitigate model fluctuations and the adverse effects of overfitting. Consequently, this enhancement led to improved training efficiency and optimal performance. Secondly, the application of differential privacy techniques ensured the privacy security in federated learning through the deliberate introduction of noise into the model gradients. Additionally, accurate quantification of privacy loss was achieved by implementing the moment accountant mechanism, facilitating a balanced trade-off between privacy preservation and analytical accuracy. This meticulous approach served to fortify system security. Lastly, the efficacy of the proposed solution was ascertained through comprehensive simulation experiments. The results substantiate the superior performance of the proposed method, evident by its exceptional accuracy, efficient utilization of privacy budget, and other notable facets.

, correspAuthors=Bao SHI, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=Chan-chan ZHAO, Kun-ming MA, Bao SHI, Xing-chen YANG, Yan LI), CN=ArticleExt(id=1245407860964835719, articleId=1245407858494390590, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=基于差分隐私的自适应联邦学习隐私保护方案, columnId=1156262729783567290, journalTitle=科学技术与工程, columnName=论文·自动化技术、计算机技术, runingTitle=null, highlight=null, articleAbstract=

随着对联邦学习的深入研究,发现联邦学习中的隐私保护策略并不能完全保护用户的隐私安全,并且在联邦学习训练过程中存在模型收敛困难的问题。针对以上问题,提出了一种自适应差分隐私机制(adaptive differential privacy, DP-AdaMod)。首先,利用自适应学习率算法调整模型训练过程,避免模型出现波动和过拟合现象,从而提高模型训练的效率和性能。其次,引入差分隐私技术,通过对模型梯度添加噪声来确保联邦学习的隐私安全。同时,使用Moment Accountant机制进行隐私损失的精确计算,有助于平衡隐私保护性能和精度,从而进一步增强了系统的安全性。最后,通过仿真实验验证所提方案的有效性。结果表明该方案在准确率、隐私预算消耗等方面展现出较优性能。

, correspAuthors=石宝, authorNote=null, correspAuthorsNote=
* 石宝(1982—),男,蒙古族,内蒙古兴安盟人,博士,教授。研究方向:人工智能。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=CnqbQv31JsOS+eTsQ+p+gw==, magXml=/RS2UagyhzpWcyWW17X6+g==, pdfUrl=null, pdf=Ur7nrioNczQGLlzSaAK2hg==, pdfFileSize=3691103, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=lY25um5isPKqxjD72inurA==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=CVbV9nW/o8TSEXmRC+hWlw==, mapNumber=null, authorCompany=null, fund=null, authors=

赵婵婵(1982—),女,汉族,山西临汾人,博士,副教授。研究方向:隐私保护、边缘计算。E-mail:

, authorsList=赵婵婵, 马坤明, 石宝, 杨星辰, 李燕)}, authors=[Author(id=1245407861346517420, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=cczhao@imut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1245407861447180725, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407861346517420, language=EN, stringName=Chan-chan ZHAO, firstName=Chan-chan, middleName=null, lastName=ZHAO, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1245407861619147193, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407861346517420, language=CN, stringName=赵婵婵, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=内蒙古工业大学信息工程学院, 呼和浩特 010080, bio={"content":"

赵婵婵(1982—),女,汉族,山西临汾人,博士,副教授。研究方向:隐私保护、边缘计算。E-mail:

"}, bioImg=null, bioContent=

赵婵婵(1982—),女,汉族,山西临汾人,博士,副教授。研究方向:隐私保护、边缘计算。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])]), Author(id=1245407861715616191, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1245407861812085187, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407861715616191, language=EN, stringName=Kun-ming MA, firstName=Kun-ming, middleName=null, lastName=MA, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1245407861887582666, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407861715616191, language=CN, stringName=马坤明, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=内蒙古工业大学信息工程学院, 呼和浩特 010080, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])]), Author(id=1245407862042771926, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=kshibao@163.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1245407862214738400, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862042771926, language=EN, stringName=Bao SHI, firstName=Bao, middleName=null, lastName=SHI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1245407862365733358, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862042771926, language=CN, stringName=石宝, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=内蒙古工业大学信息工程学院, 呼和浩特 010080, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])]), Author(id=1245407862537699827, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1245407862684500480, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862537699827, language=EN, stringName=Xing-chen YANG, firstName=Xing-chen, middleName=null, lastName=YANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1245407862801940999, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862537699827, language=CN, stringName=杨星辰, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=内蒙古工业大学信息工程学院, 呼和浩特 010080, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])]), Author(id=1245407862902604306, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1245407863045210657, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862902604306, language=EN, stringName=Yan LI, firstName=Yan, middleName=null, lastName=LI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1245407863141679658, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, authorId=1245407862902604306, language=CN, stringName=李燕, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=内蒙古工业大学信息工程学院, 呼和浩特 010080, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])])], keywords=[Keyword(id=1245407863280091705, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, orderNo=1, keyword=federated learning), Keyword(id=1245407863384949315, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, orderNo=2, keyword=differential privacy), Keyword(id=1245407863527555655, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, orderNo=3, keyword=privacy protection), Keyword(id=1245407863611441743, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, orderNo=4, keyword=adaptive), Keyword(id=1245407863716299353, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, orderNo=1, keyword=联邦学习), Keyword(id=1245407863816962655, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, orderNo=2, keyword=差分隐私), Keyword(id=1245407863926014567, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, orderNo=3, keyword=隐私保护), Keyword(id=1245407864081203823, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, orderNo=4, keyword=自适应)], refs=[Reference(id=1245407866480345945, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=1273, pageEnd=1282, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=Mcmahan H B, Moore E, Ramage D, journalName=Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, refType=null, unstructuredReference=Mcmahan H B, Moore E, Ramage D, et al. Communication-efficient learning of deep networks from decentralized data[C]// Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. Fort Lauderdale: JMLR Workshop and Conference Proceedings, 2017: 1273-1282., articleTitle=Communication-efficient learning of deep networks from decentralized data, refAbstract=null), Reference(id=1245407866589397864, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2006, volume=null, issue=null, pageStart=1, pageEnd=12, url=null, language=null, rfNumber=[2], rfOrder=1, authorNames=Dwork C, journalName=Springer Berlin Heidelberg, refType=null, unstructuredReference=Dwork C. Differential privacy[C]//International Colloquium on automata, languages, and Programming. Berlin: Springer Berlin Heidelberg, 2006: 1-12., articleTitle=Differential privacy, refAbstract=null), Reference(id=1245407866732004220, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2021, volume=21, issue=9, pageStart=3388, pageEnd=3401, url=null, language=null, rfNumber=[3], rfOrder=2, authorNames=Wei K, Li J, Ding M, journalName=IEEE Transactions on Mobile Computing, refType=null, unstructuredReference=Wei K, Li J, Ding M, et al. User-level privacy preserving federated learning: Analysis and performance optimization[J]. IEEE Transactions on Mobile Computing, 2021, 21(9): 3388-3401, articleTitle=User-level privacy preserving federated learning: Analysis and performance optimization, refAbstract=null), Reference(id=1245407866811696008, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[4], rfOrder=3, authorNames=Rü M, Kim K, journalName=arxiv preprint arxiv: 2106. 06127, 2021, refType=null, unstructuredReference= M, Kim K. Differentially private federated learning via inexact admm[J]. arxiv preprint arxiv: 2106. 06127, 2021., articleTitle=Differentially private federated learning via inexact admm, refAbstract=null), Reference(id=1245407867050771354, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=23, issue=null, pageStart=2529, pageEnd=2545, url=null, language=null, rfNumber=[5], rfOrder=4, authorNames=Huang X, Ding Y, Jiang Z L, journalName=World Wide Web, refType=null, unstructuredReference=Huang X, Ding Y, Jiang Z L, et al. DP-FL: a novel differentially private federated learning framework for the unbalanced data[J]. World Wide Web, 2020, 23: 2529-2545., articleTitle=DP-FL: a novel differentially private federated learning framework for the unbalanced data, refAbstract=null), Reference(id=1245407867205960618, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[6], rfOrder=5, authorNames=Hu R, Guo Y, Ratazzi E P, journalName=arXiv preprint arXiv: 2003. 12705, 2020, refType=null, unstructuredReference=Hu R, Guo Y, Ratazzi E P, et al. Differentially private federated learning for resource-constrained internet of things[J]. arXiv preprint arXiv: 2003. 12705, 2020., articleTitle=Differentially private federated learning for resource-constrained internet of things, refAbstract=null), Reference(id=1245407867319206840, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=25, issue=null, pageStart=2421, pageEnd=2433, url=null, language=null, rfNumber=[7], rfOrder=6, authorNames=Zhang J, Zhao Y, Wang J, journalName=Mobile Networks and Applications, refType=null, unstructuredReference=Zhang J, Zhao Y, Wang J, et al. FedMEC: improving efficiency of differentially private federated learning via mobile edge computing[J]. Mobile Networks and Applications, 2020, 25: 2421-2433., articleTitle=FedMEC: improving efficiency of differentially private federated learning via mobile edge computing, refAbstract=null), Reference(id=1245407867424064451, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=51, issue=1, pageStart=237, pageEnd=252, url=null, language=null, rfNumber=[8], rfOrder=7, authorNames=Leng J, Ye S, Zhou M, journalName=IEEET ransactions on Systems, Man, and Cybernetics: Systems, refType=null, unstructuredReference=Leng J, Ye S, Zhou M, et al. Block chain-secured smart manufacturing in industry 4. 0: a survey[J]. IEEET ransactions on Systems, Man, and Cybernetics: Systems, 2020, 51(1): 237-252., articleTitle=Block chain-secured smart manufacturing in industry 4. 0: a survey, refAbstract=null), Reference(id=1245407867541504979, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=17, pageStart=7407, pageEnd=7419, url=null, language=null, rfNumber=[9], rfOrder=8, authorNames=江欣俞, 李晓会, 秦若婷, journalName=科学技术与工程, refType=null, unstructuredReference=江欣俞, 李晓会, 秦若婷, . 基于图神经网络的兴趣点推荐的隐私保护框架[J]. 科学技术与工程, 2023, 23(17): 7407-7419., articleTitle=基于图神经网络的兴趣点推荐的隐私保护框架, refAbstract=null), Reference(id=1245407867654751200, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2023, volume=23, issue=17, pageStart=7407, pageEnd=7419, url=null, language=null, rfNumber=[9], rfOrder=9, authorNames=Jiang Xinyu, Li Xiaohui, Qin Ruoting, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Jiang Xinyu, Li Xiaohui, Qin Ruoting, et al. Privacy protection framework based on recommendation of points of interest[J]. Science Technology and Engineering, 2023, 23 (17): 7407-7419., articleTitle=Privacy protection framework based on recommendation of points of interest, refAbstract=null), Reference(id=1245407867776386027, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2024, volume=51, issue=3, pageStart=158, pageEnd=169, url=null, language=null, rfNumber=[10], rfOrder=10, authorNames=李洋, 徐进, 朱建明, journalName=西安电子科技大学学报, refType=null, unstructuredReference=李洋, 徐进, 朱建明, . 可实现双向自适应差分隐私的联邦学习方案[J]. 西安电子科技大学学报, 2024, 51(3): 158-169., articleTitle=可实现双向自适应差分隐私的联邦学习方案, refAbstract=null), Reference(id=1245407867910603770, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2024, volume=51, issue=3, pageStart=158, pageEnd=169, url=null, language=null, rfNumber=[10], rfOrder=11, authorNames=Li Yang, Xu Jin, Zhu Jianming, journalName=Journal of Xidian University, refType=null, unstructuredReference=Li Yang, Xu Jin, Zhu Jianming, et al. A federated learning scheme that can achieve bidirectional adaptive differential privacy[J]. Journal of Xidian University, 2024, 51 (3): 158-169., articleTitle=A federated learning scheme that can achieve bidirectional adaptive differential privacy, refAbstract=null), Reference(id=1245407868002877443, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1867, pageEnd=1876, url=null, language=null, rfNumber=[11], rfOrder=12, authorNames=Xu Z, Shi S, Liu A X, journalName=IEEE INFOCOM 2020-IEEE Conference on Computer Communications, refType=null, unstructuredReference=Xu Z, Shi S, Liu A X, et al. An adaptive and fast convergent approach to differentially private deep learning[C]// IEEE INFOCOM 2020-IEEE Conference on Computer Communications. New York: IEEE, 2020: 1867-1876., articleTitle=An adaptive and fast convergent approach to differentially private deep learning, refAbstract=null), Reference(id=1245407868107735057, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=8, issue=null, pageStart=23920, pageEnd=23935, url=null, language=null, rfNumber=[12], rfOrder=13, authorNames=Ye D, Yu R, Pan M, journalName=IEEE Access, refType=null, unstructuredReference=Ye D, Yu R, Pan M, et al. Federated learning in vehicular edge computing: a selective model aggregateon approach[J]. IEEE Access, 2020, 8: 23920-23935., articleTitle=Federated learning in vehicular edge computing: a selective model aggregateon approach, refAbstract=null), Reference(id=1245407868208398365, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=35, issue=4, pageStart=70, pageEnd=82, url=null, language=null, rfNumber=[13], rfOrder=14, authorNames=Liu Y, Kang Y, Xing C, journalName=IEEE Intelligent Systems, refType=null, unstructuredReference=Liu Y, Kang Y, Xing C, et al. A secure federated transfer learning framework[J]. IEEE Intelligent Systems, 2020, 35(4): 70-82., articleTitle=A secure federated transfer learning framework, refAbstract=null), Reference(id=1245407868325838891, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2014, volume=9, issue=3/4, pageStart=211, pageEnd=407, url=null, language=null, rfNumber=[14], rfOrder=15, authorNames=Dwork C, Roth A, journalName=Foundations and Trends© in Theoretical Computer Science, refType=null, unstructuredReference=Dwork C, Roth A. The algorithmic foundations of differential privacy[J]. Foundations and Trends© in Theoretical Computer Science, 2014, 9(3/4): 211-407., articleTitle=The algorithmic foundations of differential privacy, refAbstract=null), Reference(id=1245407868413919284, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2019, volume=10, issue=2, pageStart=1, pageEnd=19, url=null, language=null, rfNumber=[15], rfOrder=16, authorNames=Yang Q, Liu Y, Chen T, journalName=ACM Transactions on Intelligent systems and Technology, refType=null, unstructuredReference=Yang Q, Liu Y, Chen T, et al. Federated machine learning: concept and applications[J]. ACM Transactions on Intelligent systems and Technology, 2019, 10(2): 1-19., articleTitle=Federated machine learning: concept and applications, refAbstract=null), Reference(id=1245407868535554109, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2018, volume=271, issue=3, pageStart=808, pageEnd=817, url=null, language=null, rfNumber=[16], rfOrder=17, authorNames=Mercier Q, Poirion F, Désidéri J A, journalName=European Journal of Operational Research, refType=null, unstructuredReference=Mercier Q, Poirion F, Désidéri J A. A stochastic multiple gradient descent algorithm[J]. European Journal of Operational Research, 2018, 271(3): 808-817., articleTitle=A stochastic multiple gradient descent algorithm, refAbstract=null), Reference(id=1245407868636217418, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[17], rfOrder=18, authorNames=Kingma D P, journalName=arxiv preprint arxiv: 1412. 6980, 2014, refType=null, unstructuredReference=Kingma D P. Adam: A method for stochastic optimization[J]. arxiv preprint arxiv: 1412. 6980, 2014., articleTitle=Adam: A method for stochastic optimization, refAbstract=null), Reference(id=1245407868728492114, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2016, volume=48, issue=4, pageStart=522, pageEnd=534, url=null, language=null, rfNumber=[18], rfOrder=19, authorNames=Fang W, Yao X, Zhao X, journalName=IEEE Transactions on Systems, Man, and Cybernetics: Systems, refType=null, unstructuredReference=Fang W, Yao X, Zhao X, et al. A stochastic control-approach to maximize profit on service provisioning for mobile cloudlet platforms[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2016, 48(4): 522-534., articleTitle=A stochastic control-approach to maximize profit on service provisioning for mobile cloudlet platforms, refAbstract=null), Reference(id=1245407868845932641, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2020, volume=21, issue=1, pageStart=9047, pageEnd=9076, url=null, language=null, rfNumber=[19], rfOrder=20, authorNames=Ward R, Wu X, Bottou L, journalName=The Journal of Machine Learning Research, refType=null, unstructuredReference=Ward R, Wu X, Bottou L. Adagrad stepsizes: Sharp convergence over nonconvex landscapes[J]. The Journal of Machine Learning Research, 2020, 21(1): 9047-9076., articleTitle=Adagrad stepsizes: Sharp convergence over nonconvex landscapes, refAbstract=null), Reference(id=1245407868925624424, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[20], rfOrder=21, authorNames=Kurbiel T, Khaleghian S, journalName=arXiv preprint arXiv: 1708. 01911, 2017, refType=null, unstructuredReference=Kurbiel T, Khaleghian S. Training of deep neural networks based on distance measures using RMSProp[J]. arXiv preprint arXiv: 1708. 01911, 2017., articleTitle=Training of deep neural networks based on distance measures using RMSProp, refAbstract=null), Reference(id=1245407868992733298, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=[21], rfOrder=22, authorNames=Ding J, Ren X, Luo R, journalName=arXiv preprint arXiv: 1910. 12249, 2019, refType=null, unstructuredReference=Ding J, Ren X, Luo R, et al. An adaptive and momental bound method for stochastic learning[J]. arXiv preprint arXiv: 1910. 12249, 2019., articleTitle=An adaptive and momental bound method for stochastic learning, refAbstract=null), Reference(id=1245407869089202300, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=308, pageEnd=318, url=null, language=null, rfNumber=[22], rfOrder=23, authorNames=Abadi M, Chu A, Goodfellow I, journalName=Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, refType=null, unstructuredReference=Abadi M, Chu A, Goodfellow I, et al. Deep learningwith differential privacy[C]// Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security. New York: ACM, 2016: 308-318., articleTitle=Deep learningwith differential privacy, refAbstract=null)], funds=[Fund(id=1245407865951863574, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, awardId=NJZY22382, language=CN, fundingSource=内蒙古自治区高等学校科学研究项目(NJZY22382), fundOrder=null, country=null), Fund(id=1245407866056721191, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, awardId=JY20240010, language=CN, fundingSource=内蒙古自治区直属高校基本科研业务费项目(JY20240010), fundOrder=null, country=null), Fund(id=1245407866132218672, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, awardId=JY20230082, language=CN, fundingSource=内蒙古自治区直属高校基本科研业务费项目(JY20230082), fundOrder=null, country=null), Fund(id=1245407866249659199, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, awardId=2023LHMS06016, language=CN, fundingSource=内蒙古自治区自然科学基金(2023LHMS06016), fundOrder=null, country=null), Fund(id=1245407866354516813, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, awardId=BS201936, language=CN, fundingSource=内蒙古工业大学学科学研究项目(BS201936), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1245407861203911068, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, xref=null, ext=[AuthorCompanyExt(id=1245407861216493981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China), AuthorCompanyExt(id=1245407861224882590, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, companyId=1245407861203911068, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=内蒙古工业大学信息工程学院, 呼和浩特 010080)])], figs=[ArticleFig(id=1245407864257364604, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=Fig.1, caption=Federated learning framework, figureFileSmall=KS48tFmuQ66lC6DDkd2qZg==, figureFileBig=lY25um5isPKqxjD72inurA==, tableContent=null), ArticleFig(id=1245407864437719689, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=图1, caption=联邦学习框架, figureFileSmall=KS48tFmuQ66lC6DDkd2qZg==, figureFileBig=lY25um5isPKqxjD72inurA==, tableContent=null), ArticleFig(id=1245407864727126690, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=Fig.2, caption=Localization of differential privacy, figureFileSmall=tmSBd0xGfyy45ETLsWhERg==, figureFileBig=afU2joVtwz3oDjLngHpehQ==, tableContent=null), ArticleFig(id=1245407864869733036, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=图2, caption=本地化差分隐私, figureFileSmall=tmSBd0xGfyy45ETLsWhERg==, figureFileBig=afU2joVtwz3oDjLngHpehQ==, tableContent=null), ArticleFig(id=1245407864987173557, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=Fig.3, caption=Overall framework, figureFileSmall=RC9sFehvUKYJBtb4hpQx2Q==, figureFileBig=A6G9D8ZbkGdYLjZPvkeHWg==, tableContent=null), ArticleFig(id=1245407865121391298, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=图3, caption=总体框架, figureFileSmall=RC9sFehvUKYJBtb4hpQx2Q==, figureFileBig=A6G9D8ZbkGdYLjZPvkeHWg==, tableContent=null), ArticleFig(id=1245407865243026126, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=Fig.4, caption=Comparison of accuracy under different privacy budget conditions, figureFileSmall=AyeuYPp80tWEi9Yq3ky/Pg==, figureFileBig=oXHB8IeXckYalSel6/HMfQ==, tableContent=null), ArticleFig(id=1245407865410798301, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=图4, caption=不同隐私预算条件下准确率对比, figureFileSmall=AyeuYPp80tWEi9Yq3ky/Pg==, figureFileBig=oXHB8IeXckYalSel6/HMfQ==, tableContent=null), ArticleFig(id=1245407865536627434, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=null, caption=null, figureFileSmall=null, figureFileBig=null, tableContent=
算法1 自适应差分隐私算法(DP-AdaMod算法)
nput:初始化参数θ0,初始化参数样本{x1,x2,…,xN},学习$\{{\eta }_{t}{\}}_{t=1}^{T}$,噪声参数σ,批次大小L,衰减参数{β1,β2,β3},梯度裁剪阈值C,迭代次数T,损失函数L(θ)=$\frac{1}{N}{\sum }_{i}^{}$L(θ,xi)。
Initialize: m0=0,v0=0,s0=0,
for t=1 to T do:
Step1:计算梯度gt(xi)← L(θ,xi)
Step2:梯度修剪gt(xi)←$\frac{{g}_{t}\left({x}_{i}\right)}{max(1,\Vert {g}_{t}({x}_{i}){\Vert }_{2}/C)}$
Step3:计算学习率和梯度累计值:${\overline{g}}_{t}$$\sum _{i=0}^{m}$gt(xi)mt=β1
mt-1+(1-β1)${\overline{g}}_{t}$
vt=β2vt-1+(1-β2)${{\overline{g}}_{t}}^{2}$
${\stackrel{\wedge }{m}}_{t}$=mt/(1-${\beta }_{1}^{t}$)
${\stackrel{\wedge }{v}}_{t}$=vt/(1-${\beta }_{2}^{t}$)
ηt=αt/($\sqrt[ ]{{\stackrel{\wedge }{v}}_{t}}$+ε)
st=β3st-1+(1-β3)ηt
${\stackrel{\wedge }{\eta }}_{t}$=min(ηt,st)
Step4:噪声添加${\stackrel{~}{g}}_{t}$$\frac{1}{L}{\sum }_{i}^{}$[${\stackrel{\wedge }{m}}_{t}$+N(0,${\sigma }_{t}^{2}{C}^{2}$I)]
Step5:参数更新θt=θt-1-${\stackrel{\wedge }{\eta }}_{t}{\stackrel{~}{g}}_{t}$
end for
Out Put: θt
), ArticleFig(id=1245407865637290741, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=null, caption=null, figureFileSmall=null, figureFileBig=null, tableContent=
算法1 自适应差分隐私算法(DP-AdaMod算法)
nput:初始化参数θ0,初始化参数样本{x1,x2,…,xN},学习$\{{\eta }_{t}{\}}_{t=1}^{T}$,噪声参数σ,批次大小L,衰减参数{β1,β2,β3},梯度裁剪阈值C,迭代次数T,损失函数L(θ)=$\frac{1}{N}{\sum }_{i}^{}$L(θ,xi)。
Initialize: m0=0,v0=0,s0=0,
for t=1 to T do:
Step1:计算梯度gt(xi)← L(θ,xi)
Step2:梯度修剪gt(xi)←$\frac{{g}_{t}\left({x}_{i}\right)}{max(1,\Vert {g}_{t}({x}_{i}){\Vert }_{2}/C)}$
Step3:计算学习率和梯度累计值:${\overline{g}}_{t}$$\sum _{i=0}^{m}$gt(xi)mt=β1
mt-1+(1-β1)${\overline{g}}_{t}$
vt=β2vt-1+(1-β2)${{\overline{g}}_{t}}^{2}$
${\stackrel{\wedge }{m}}_{t}$=mt/(1-${\beta }_{1}^{t}$)
${\stackrel{\wedge }{v}}_{t}$=vt/(1-${\beta }_{2}^{t}$)
ηt=αt/($\sqrt[ ]{{\stackrel{\wedge }{v}}_{t}}$+ε)
st=β3st-1+(1-β3)ηt
${\stackrel{\wedge }{\eta }}_{t}$=min(ηt,st)
Step4:噪声添加${\stackrel{~}{g}}_{t}$$\frac{1}{L}{\sum }_{i}^{}$[${\stackrel{\wedge }{m}}_{t}$+N(0,${\sigma }_{t}^{2}{C}^{2}$I)]
Step5:参数更新θt=θt-1-${\stackrel{\wedge }{\eta }}_{t}{\stackrel{~}{g}}_{t}$
end for
Out Put: θt
), ArticleFig(id=1245407865733759745, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=EN, label=Table 2, caption=

Comparison of privacy budget consumption under the same accuracy condition

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集 准确率/
%
δ ε1 ε2 ε3 减少率
1/%
减少率
2/%
MNIST 88 10-5 0.71 0.62 0.56 21.13 8.94
90 1.28 1.09 0.92 28.04 15.51
92 1.78 1.32 1.23 30.90 6.81
94 5.73 3.68 2.98 47.99 19.02
), ArticleFig(id=1245407865838617355, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1245407858494390590, language=CN, label=表2, caption=

相同准确率条件下消耗隐私预算对比

, figureFileSmall=null, figureFileBig=null, tableContent=
数据集 准确率/
%
δ ε1 ε2 ε3 减少率
1/%
减少率
2/%
MNIST 88 10-5 0.71 0.62 0.56 21.13 8.94
90 1.28 1.09 0.92 28.04 15.51
92 1.78 1.32 1.23 30.90 6.81
94 5.73 3.68 2.98 47.99 19.02
)], attaches=null, journal=Journal(id=1146119176004939786, delFlag=0, nameCn=科学技术与工程, nameEn=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, issn=1671-1815, eissn=, cn=11-4688/T, coden=null, periodic=4, language=CN, oaType=是, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=UKU/O7GSka5polgCTkbIIw==, journalPrice=null, startedYear=null, abbrevIsoEn=Sci Technol Eng, journalRemark=null, publicationField=null, createdTime=null, updatedTime=1754445529766, createdBy=null, updatedBy=13701087609, firstLetterCn=S, firstLetterEn=S, subjectCode=Natural Sciences, subjectName=自然科学, subjectCodeEn=Natural Sciences, subjectNameEn=null, picCn=UKU/O7GSka5polgCTkbIIw==, picEn=5hwlULoNwcbj3xUmVi9MAQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1159791870395564357, language=CN, name=科学技术与工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529793, updatedTime=1754445529793, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=http://www.stae.com.cn/jsygc/site/menus/20090429150146001, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1159791870441701702, language=EN, name=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529804, updatedTime=1754445529804, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1146123166801305609, websiteList=[Website(id=1148243202391400884, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/CN, language=CN, createTime=1751692112777, createBy=18614031015, updateTime=1753520965431, updateBy=18614031015, name=科学技术与工程-中文站点, tplId=1146099689490845704, title=科学技术与工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1148622798802673703, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=articleTextType, value=kx, createTime=1751782615614, updateTime=1751782615614, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798781702180, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=banner, value=null, createTime=1751782615609, updateTime=1751782615609, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798769119267, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1751782615606, updateTime=1751782615606, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798794285094, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1751782615612, updateTime=1751782615612, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798790090789, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1751782615611, updateTime=1751782615611, creator=18614031015, updator=18614031015)]), Website(id=1155914124811976731, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/EN, language=EN, createTime=1753521003206, createBy=18614031015, updateTime=1753521003206, updateBy=18614031015, name=科学技术与工程-英文站点, tplId=1146101810881728533, title=Science Technology and Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1155914371227308235, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=articleTextType, value=kx, createTime=1753521061952, updateTime=1753521061952, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371210531016, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=banner, value=null, createTime=1753521061947, updateTime=1753521061947, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371202142407, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1753521061945, updateTime=1753521061945, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371223113930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1753521061950, updateTime=1753521061950, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371218919625, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1753521061949, updateTime=1753521061949, creator=18614031015, updator=18614031015)])], journalTitle=科学技术与工程, weixinUrl=null, journalUrl=null, iacademicId=null, status=0, seqNo=null, journalTitleEn=Science Technology and Engineering, journalPhotoCn=UKU/O7GSka5polgCTkbIIw==, journalPhotoEn=5hwlULoNwcbj3xUmVi9MAQ==, journalFirstLetter=S, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=null, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2308502, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2308502, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2308502, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2308502, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
基于差分隐私的自适应联邦学习隐私保护方案
收藏切换
PDF下载
赵婵婵 , 马坤明 , 石宝 * , 杨星辰 , 李燕
科学技术与工程 | 论文·自动化技术、计算机技术 2025,25(7): 2849-2855
收起
收藏切换
科学技术与工程 | 论文·自动化技术、计算机技术 2025, 25(7): 2849-2855
基于差分隐私的自适应联邦学习隐私保护方案
全屏
赵婵婵 , 马坤明, 石宝* , 杨星辰, 李燕
作者信息
  • 内蒙古工业大学信息工程学院, 呼和浩特 010080
  • 赵婵婵(1982—),女,汉族,山西临汾人,博士,副教授。研究方向:隐私保护、边缘计算。E-mail:

通讯作者:

* 石宝(1982—),男,蒙古族,内蒙古兴安盟人,博士,教授。研究方向:人工智能。E-mail:
Adaptive Privacy Protection Scheme for Federated Learning Based on Differential Privacy
Chan-chan ZHAO , Kun-ming MA, Bao SHI* , Xing-chen YANG, Yan LI
Affiliations
  • College of Information Science and Engineering, Inner Mongolia University of Technology, Hohhot 010080, China
出版时间: 2025-03-08 doi: 10.12404/j.issn.1671-1815.2308502
文章导航
收藏切换

随着对联邦学习的深入研究,发现联邦学习中的隐私保护策略并不能完全保护用户的隐私安全,并且在联邦学习训练过程中存在模型收敛困难的问题。针对以上问题,提出了一种自适应差分隐私机制(adaptive differential privacy, DP-AdaMod)。首先,利用自适应学习率算法调整模型训练过程,避免模型出现波动和过拟合现象,从而提高模型训练的效率和性能。其次,引入差分隐私技术,通过对模型梯度添加噪声来确保联邦学习的隐私安全。同时,使用Moment Accountant机制进行隐私损失的精确计算,有助于平衡隐私保护性能和精度,从而进一步增强了系统的安全性。最后,通过仿真实验验证所提方案的有效性。结果表明该方案在准确率、隐私预算消耗等方面展现出较优性能。

联邦学习  /  差分隐私  /  隐私保护  /  自适应

With the deepening research on federated learning, it has been observed that the privacy protection strategies employed within federated learning fall short of fully guaranteeing the security and confidentiality of user data. Moreover, the training process in federated learning encounters challenges regarding model convergence. In response to these aforementioned issues, an innovative solution termed adaptive differential privacy (DP-AdaMod) was proposed. Primarily, the model training process was fine-tuned by incorporating an adaptive learning rate algorithm to mitigate model fluctuations and the adverse effects of overfitting. Consequently, this enhancement led to improved training efficiency and optimal performance. Secondly, the application of differential privacy techniques ensured the privacy security in federated learning through the deliberate introduction of noise into the model gradients. Additionally, accurate quantification of privacy loss was achieved by implementing the moment accountant mechanism, facilitating a balanced trade-off between privacy preservation and analytical accuracy. This meticulous approach served to fortify system security. Lastly, the efficacy of the proposed solution was ascertained through comprehensive simulation experiments. The results substantiate the superior performance of the proposed method, evident by its exceptional accuracy, efficient utilization of privacy budget, and other notable facets.

federated learning  /  differential privacy  /  privacy protection  /  adaptive
赵婵婵, 马坤明, 石宝, 杨星辰, 李燕. 基于差分隐私的自适应联邦学习隐私保护方案. 科学技术与工程, 2025 , 25 (7) : 2849 -2855 . DOI: 10.12404/j.issn.1671-1815.2308502
Chan-chan ZHAO, Kun-ming MA, Bao SHI, Xing-chen YANG, Yan LI. Adaptive Privacy Protection Scheme for Federated Learning Based on Differential Privacy[J]. Science Technology and Engineering, 2025 , 25 (7) : 2849 -2855 . DOI: 10.12404/j.issn.1671-1815.2308502
新兴的技术如云计算、边缘计算和物联网的发展导致全球数据量急剧增长,这给数据的流通和共享带来了巨大的挑战,特别是数据安全方面的威胁变得日益严峻。
传统中心化机器学习通过将数据集中上传至中央服务器进行模型训练,这会导致数据传输过程容易被攻击者截获,造成隐私泄露问题。McMahan等[1]在2016年首次提出联邦学习(federated learning, FL)的概念,并于2017年提出联邦平均算法(federal average algorithm,FedAvg)。联邦学习是一种分布式机器学习方法,旨在实现在不将参与方的私有数据上传到中央服务器的情况下,以隐私保护的方式交换中间模型参数,从而协作完成分布式任务。当前,联邦学习的研究主要集中在如何保护隐私和确保安全性方面,而差分隐私技术是一个重要的隐私保护手段,其采用对数据添加噪声的方法来保护隐私。自从2016年Dwork[2]首次提出差分隐私技术(differential privacy,DP)的概念以来,研究者们根据这一概念进行了大量的研究,提出了许多满足差分隐私要求的算法。Wei等[3]为了增强联邦学习的隐私保护性能,提出了一种本地化差分隐私机制,该方法通过在每个联邦学习参与方本地添加噪声实现隐私保护,并分析了在相同隐私预算消耗的情况下,参与方数量与全局模型收敛之间的关系。Rü等[4]在研究中将联邦学习与差分隐私技术相结合,提出了一种差分隐私模糊交替方向乘子算法DP-IADMM(differential privacy inexact alternating direction method of multipliers),该算法解决了随机噪声对目标扰动的子问题。Huang等[5]针对联邦学习数据分布不平衡的问题,提出了一种差分隐私联邦学习框架(differential privacy federated learning,DP-FL),通过使用新的模型聚合算法,引入权重调整机制以及应用差分隐私机制保护隐私,从而确保在不平衡数据分布情况下的有效性。Hu等[6]提出了一种基于差分隐私技术的轻量级联邦学习模型聚合方法DP-PASGD(differential privacy periodic averaging SGD),通过引入差分隐私技术,提升了隐私保护性能,并通过优化参数聚合过程,降低了通信和计算开销。Zhang等[7]提出了一种基于移动边缘计算的联邦学习框架(federated learning mobile edge computing,FedMEC),通过将部分计算任务迁移到边缘服务器上,并引入差分隐私技术,提高了差分隐私联邦学习的效率和隐私保护性能。Leng等[8]使用了许多优化算法,改善由于联邦学习模型固定的学习率导致收敛过程出现局部收敛的情况。江欣俞等[9]提出了一种基于图神经网络的兴趣点推荐的隐私保护框架,并采用可变动态梯度的客户端差分隐私算法,达到了边优化边反馈的效果。李洋等[10]提出一种可实现双向自适应差分隐私的联邦学习方案(federated learning mobile bidirectional adaptive differential privacy,FedBADP)可以对数据之间传输的梯度进行自适应加噪,但在同步联邦学习训练方式下,仍存在参数泄露的风险。Xu等[11]通过使用自适应梯度下降来加速模型收敛和降低隐私损失,但该方案的模型准确率较低。
综上所述,当前差分隐私与联邦学习的结合,虽然对联邦学习的隐私保护能力有了很大提升,但是由于噪声会对模型的精度造成影响,很难在模型的隐私性和准确性达到一个平衡。为此,提出一种自适应差分隐私优化算法,可以自适应调整学习率,从而有效降低差分隐私添加噪声对模型精度的影响。本文的重要工作如下。
(1)提出了一种自适应差分隐私优化算法。它可以根据历史梯度信息自适应地调整学习率,以更好地适应不同参数的变化。这使得在联邦学习中,模型能够更好地适应不同客户端的数据特点,提高全局模型的准确性。
(2)引入差分隐私保护机制,确保在联邦学习过程中隐私得到保护。通过应用差分隐私技术,对模型更新的梯度添加噪声,从而保护具体客户端的隐私。这使得本文方案可以在联邦学习中更好地平衡模型精度和隐私保护需求。
联邦学习(FL)是一种分布式机器学习方法,可以根据参与方数据集的特征分为以下3类:横向联邦学习、纵向联邦学习和联邦迁移学习[12-13]
图1为典型联邦学习系统框架,由多个参与方和一个聚合服务器组成。每个参与方都持有数据特征完整的且数据样本之间几乎没有交集的本地数据集。通过聚合服务器的协调,参与方可以联合起来训练一个更加高效的全局模型。具体训练过程如下。
(1)模型初始化:联邦学习开始之前,需要初始化一个全局模型。聚合服务器初始化全局模型并选择适合于训练的客户端。
(2)数据分发:服务器将初始化的全局模型发送给参与方用作本地训练的初始模型。
(3)本地训练:参与方接收到全局模型后,使用本地数据集训练本地模型。
(4)模型聚合:每个参与方完成本地训练后,聚合服务器收集各参与方的模型参数并使用聚合算法进行模型聚合。
(5)模型更新:聚合服务器将聚合后的全局模型发送给所有参与方,用于更新本地模型。
(6)重复迭代:重复步骤(2)~(5),直至全局模型收敛或者达到预定停止条件。
差分隐私是一种隐私保护技术,旨在通过在原始数据集或数据交换过程中添加精心设计的噪声,以保护数据隐私。差分隐私可以分为中心化差分隐私和本地化差分隐私[2]。中心化差分隐私是指在集中式环境中,通过在中央服务器对数据添加噪声以保护数据隐私。本地化差分隐私则是指在分布式环境中,将数据处理过程全部在本地设备完成,用户可以在本地对数据进行加噪处理以保护敏感信息。本文中主要应用本地化差分隐私思想,本地化差分隐私的实现如图2所示。
定义1[14] (ε,δ)-差分隐私。对于给定ε>0和δ>0,若其算法M对于任意相邻数据集 DD',其输出结果S满足关系式
Pr[M(D)∈S]≤eεPr[M(D')∈S]+δ
则证明M满足(ε,δ)-差分隐私。其中,ε为隐私预算,δ是一个常数项,用来衡量隐私保护机制中允许的随机性额外噪声的影响。
定义2[14] 隐私敏感度。对于任意查询函数f,将数据集D映射到输出空间中,隐私敏感度可以表示为
Δf=$\underset{D,D\text{'}}{max}$f(D)-f(D')‖p
式(2)中: ‖·‖p表示Lp范数,即对于 DD',函数f的最大变化值。隐私敏感度是差分隐私中的关键定义,用来衡量或者量化查询函数fDD'上输出结果变化的敏感程度。隐私敏感度越高,则隐私泄露的风险也会越大。为了降低隐私泄露风险,添加的噪声也会相应变大。
定义3[14] 高斯噪声。对于查询函数f,向其输出f(D)中添加高斯噪声,实现(ε,δ)——差分隐私。即
M(D)=f(D)+N[0,(Δfσ)2I]
式(3)中:高斯分布的均值为0;协方差为(Δfσ)2I,I为单位矩阵。
学习率和迭代次数是机器学习过程之前需要设置的超参数。传统随机梯度下降算法(stochastic gradient descent,SGD)[15-16]由于固定的学习率,往往会在训练过程中减慢收敛速度并且有很强的震荡性。为了更好地适应不同维度目标函数的优化过程,自适应学习率的梯度下降算法开始被研究者们研究和使用。Adam[17-18]兼具了AdaGrad[19]和RMSProp[20]的优点,它也可以利用一阶矩均值计算自适应学习率;不同的是,Adam还充分利用了梯度的二阶矩均值。然而,尽管Adam算法具备这些优点,但是Adam仍然存在着在某些情况下不收敛和波动大的情况。AdaMod优化算法[21]是基于Adam算法的思想进行改进的,AdaMod算法作为一种自适应学习率的算法,能够在训练开始时就控制自适应学习率的方差,以确保训练初期的稳定性。因此,在AdaMod优化算法思想基础上进行改进,提出了一种自适应差分隐私优化算法(DP-AdaMod),通过在梯度中加入噪声保护隐私,并利用自适应学习率加快模型收敛,有效平衡了模型的隐私性和准确性。
AdaMod优化算法计算一阶矩的步骤如下:
(1)在每轮迭代过程中,算法会从训练集中随机选出m个样本,然后通过计算先前一阶矩的指数移动平均值计算当前梯度的一阶矩。即
mt=β1mt-1+(1-β1)gt
式(4)中:mt为第t次迭代时梯度的一阶矩估计值;β1为衰减率,其值基于0和1;gt为梯度。
(2)AdaMod算法通过使用梯度gt平方的指数移动平均数来估算梯度的二阶矩。具体来说,在每次迭代过程中, AdaMod优化算法通过计算先前二阶矩的指数移动平均值来估计当前梯度的二阶矩。该估算值将用于调整学习率,以适应梯度变化。初始时,二阶矩估计为零向量。然后计算进行二阶矩,公式为
vt=β2vt-1+(1-β2)${g}_{t}^{2}$
式(5)中:β2为衰减率;vt为第t次迭代时梯度平方的二阶矩估计值。通过计算一阶矩和二阶矩,AdaMod算法能够自适应地调整梯度更新步长,并在训练中动态地平衡参数更新的速度和方向,以提高训练效果。
(3)AdaMod算法通过引入修正项以减少初始迭代时估计偏差的影响。具体而言,在每次迭代中,将${\beta }_{1}^{t}$${\beta }_{2}^{t}$分别用于一阶矩和二阶矩的估算,得到修正后的一阶矩${\stackrel{\wedge }{m}}_{t}$和二阶矩${\stackrel{\wedge }{v}}_{t}$
$\left\{\begin{array}{l}{\stackrel{\wedge }{m}}_{t}=\frac{{m}_{t}}{1-{\beta }_{1}^{t}}\\ {\stackrel{\wedge }{v}}_{t}=\frac{{v}_{t}}{1-{\beta }_{2}^{t}}\end{array}\right.$
通过引入修正量,AdaMod算法能够减少初始迭代时估计偏差的影响,从而更准确地估计一阶矩和二阶矩,提高算法在初始阶段的稳定性和性能。
(4)最后,在每一轮的迭代中,参数值θt将通过学习率ηt${\stackrel{\wedge }{m}}_{t}$来计算,即
θt+1θt-ηt${\stackrel{\wedge }{m}}_{t}$
AdaMod算法通过对学习率进行指数加权平均来计算当前指数的滑动平均值,即
st=β3st-1+(1-β3)ηt
st=(1-β3)[st-1+β3st-2+…+${\beta }_{3}^{t-1}$s0]
式中:st为指数滑动平均值;β3为记忆时长参数,其值小于1。
根据式(9),可以计算出当前的指数滑动平均值。然后,比较当前指数滑动平均值st和学习率ηt的值,将较小的值作为当前的学习率。这样做有效避免了学习率过高的问题。通过不断重复上述步骤,可以保证学习率一直受到当前指数滑动平均值的限制,使其更稳定地适应模型训练需求。
DP-AdaMod算法的主要步骤包括梯度剪裁、学习率计算和添加噪声。梯度剪裁是一种常见的模型优化技术,限制梯度范围在适当的范围内,以防止梯度爆炸问题,并加速模型的收敛。在DP-AdaMod算法中,对每个梯度向量的L2范数应用梯度剪裁。梯度剪裁的公式为
gt(xi)←$\frac{{g}_{t}\left({x}_{i}\right)}{max\left[1\right.,\Vert {g}_{t}\left({x}_{i}\right){\Vert }_{2}/C]}$
式(10)中: gt为当前梯度;C为裁剪阈值。
噪声添加的公式为
$\tilde{g}_{t} \leftarrow \frac{1}{L} \sum_{i}\left[\hat{m}_{t}+N\left(0, \sigma_{t}^{2} C^{2} \boldsymbol{I}\right)\right]$
本文中使用高斯噪声N(0,σ2C2I),噪声的概率分布是标准差为、均值为0的高斯分布。对于梯度向量 gt,如果其L2范数$\Vert {g}_{t}{\Vert }_{2}$C,则保留原始梯度gt;如果其L2范数$\Vert {g}_{t}{\Vert }_{2}$>C时,则将梯度裁剪为C,即gt=C。因此,可以通过设置裁剪阈值C的值调整后续添加的噪声大小,这意味着后续添加的噪声大小与裁剪阈值C的大小密切相关。这种方式允许能够对梯度向量进行裁剪,并且能够对噪声大小进行控制,从而实现隐私保护性能和模型精度之间的平衡。
DP-AdaMod算法计算学习率的过程为:首先,计算一阶矩和二阶矩的估计值。其次,使用一阶矩的估计值和二阶矩的估计值来计算出学习率ηt
然后,计算当前的指数滑动平均值st。最后对比ηtst值的大小,将二者中值较小的作为当前学习率。
本文算法的具体实现如下:
中心服务器并不会收集所有客户端的数据,而是通过收集梯度更新来更新中心模型,此过程可以在一定程度上防止攻击者反推出模型数据。对于每个参与计算的客户端,每轮前都会从服务器获取最新的模型参数作为初始模型参数,并利用本地模型计算梯度更新并发送给中心服务器。中心模型的更新过程主要包括如下几个关键步骤。
(1)中央服务器选择适合于训练的客户端,同时发送初始参数模型并开放下载权限。
(2)参与方使用本地数据集更新参数并计算更新的梯度。在这一步骤中使用自适应梯度下降算法来保证良好收敛性。
(3)避免单个节点过度贡献,对一些参与方更新的梯度进行缩放,使模型可以更好地综合各参与方的特征,从而提高整体模型性能。
(4)将差分隐私噪声添加到单个通信过程中的整体聚合更新中。
(5)将梯度更新的结果发送到中央服务器,同时,将更新的全局模型广播给参与方。图3为本文中总体架构。
在DP-AdaMod算法中,每进行一次梯度计算或者更新一次参数都会产生隐私损失。因此,在一定隐私预算条件下,能够准确计算算法的隐私损失尤为重要。本文中利用Abadi等[22]提出的时刻统计方法(moment accountant)来进行隐私统计。该方法可以帮助参与方更好地平衡隐私性和准确率,可以增强系统的安全性。相关定义如下:
定义4 隐私损失[22]。对于随机算法M,其隐私损失是通过计算M在相邻数据集DD'上产生相同输出概率之比的对数来表示,即
$c\left(S ; M, A, D, D^{\prime}\right) \triangleq \lg \frac{\operatorname{Pr}[M(A, D)=S]}{\operatorname{Pr}\left[M\left(A, D^{\prime}\right)=S\right]}$
式(12)中:A为辅助变量。
定义5 时刻统计[22]。算法Mλ时刻的时刻统计为
$\alpha(\lambda) \triangleq \max _{D, D^{\prime}} \lg E_{S \sim M(D)}\left\{\exp \left[\lambda c\left(S, M, D, D^{\prime}\right)\right]\right\}$
定理1 组合特性[22]。给定算法M和一系列子算法M1,M2,…,Mn,其对任意时刻λ的时刻统计上界等于所有子算法在该时刻的时刻统计之和。
αM(λ)≤$\sum _{i=1}^{N}{\alpha }_{{M}_{i}}$(λ)
式(14)中:αM(λ)为算法Mλ时刻的时刻统计上届;αMi(λ)为子算法Miλ时刻的时刻统计上界。
定理2 差分隐私边界[22]。对于任意ε>0,若满足
δ=$\underset{\lambda }{min}$exp[αM(λ)-λε]
则算法M满足(ε,δ)-差分隐私。
根据定义5计算得到DP-AdaMod算法在λ时刻统计的值为$\alpha_{M}\left(\lambda ; A, D, D^{\prime}\right) \triangleq \max _{D, D^{\prime}} \lg E_{S \sim M(A, D)} \left\{\exp \left[\lambda c\left(S ; M, A, D, D^{\prime}\right)\right]\right\} $。
根据组合特性,假设算法的时刻统计为α(λ),可以推出
α(λ)≤$\sum _{t}^{T}\sum _{i=1}^{N}$αi,j(λ)
式(16)中:αi,j(λ)在梯度上添加高斯噪声ξ~N(0,σ2C2I);T为总训练轮数;N为设备总数。TN与DP-AdaMod算法的隐私损失成正比。αi,j(λ)的计算过程如下。
μ0表示高斯分布N(0,σ2C2I)的概率密度函数,μ1表示高斯分布N(1,σ2C2I)的概率密度函数,μ表示同时混合μ0μ1的高斯分布,即μ=(1-q)μ0+1,其中q为进行本地训练时的抽样概率。 根据式(12)和式(13)可以推出αi,j(λ)的表达式αi,j(λ)=lg$\underset{}{max}$(E1,E2),其中E1E2可以分别表示为E1=Ex-μ0{$\left[{\mu }_{0}\right(x)/\mu {\left(x\right)]}^{2}$},E2=Ex-μ{$\left[\mu \right(x)/{\mu }_{0}{\left(x\right)]}^{2}$}。根据上述推导可得隐私损失边界为αi,j(λ)≤q2λ(λ+1)/(1-q)σ2+O(q33)。同时,因为本地设备添加的$噪声\xi ~N(0,{\sigma }^{2}{C}^{2}$I)相同,根据计算得到的α(λ),利用定理2推出满足(ε,δ)=$\underset{\lambda }{min}$exp[α(λ)-λε]-差分隐私。
本文中网络结构采用LeNet-5卷积神经网络。由3个5×5的卷积层(C1、C3、C5),两个2×2的池化层(S2、S4)和一个输出层组成,输出层共有10个神经元,输出为0~9。实验采用MNIST数据集,该数据集被广泛应用于联邦学习的测试中。MNIST是一个手写数字识别数据集,包含大约70 000张28×28的灰度图像,标签取值为0~9。该数据集由训练集和测试集两部分组成。其中,训练集包含60 000张训练图像,用于训练模型参数;测试集包含10 000个测试图像,用于评估模型性能。
为了评估DP-AdaMod算法的准确率,在隐私预算分别为ε=0.5,2,8条件下,将DP-AdaMod算法与DPSGD和DPAdam算法进行准确率的比较。实验设置梯度裁剪阈值为C=1、噪声参数σ=4、初始学习率η=0.001。结果如图4所示。
图4所示,在隐私预算ε的值分别为0.5、2和8的条件下,提出的DP-AdaMod算法均取得了不错的效果,准确率优于DPSGD和DPAdam算法。实验结果表明,在引入差分隐私技术的联邦学习训练过程中,与固定学习率的梯度下降算法相比,DP-AdaMod算法能够更好地平衡隐私性和准确性。本文算法可以有效防止学习率过大情况,使得每一次学习率自适应更新的值更加精确,保证联邦学习在加噪场景依旧可以保持较好的性能。
为了衡量DP-AdaMod算法在减少隐私预算消耗方面的作用,采用DPSGD和DPAdam作为对比。表2是在准确率分别为88%、90%、92%和94%时上述3种算法的隐私预算消耗。其中ε1ε2ε3分别代表DP-SGD、DP-Adam和DP-AdaMod3种算法消耗的隐私预算,减少率1代表DP-AdaMod算法相对于DP-SGD隐私预算消耗减少率,减少率2代表DP-AdaMod算法相对于DP-Adam算法隐私预算消耗减少率。
表2可以看出,DP-AdaMod算法在MNIST数据集上相比DP-SGD算法平均减少了约32%的隐私预算,比DP-Adam算法平均减少了12.57%的隐私预算。结果表明,DP-AdaMod算法在达到相同准确率的条件下,减少了隐私预算消耗。随着隐私预算的增加,隐私保护程度会降低。因此,在实现相同准确率的前提下,与其他两种算法相比,DP-AdaMod算法的隐私预算消耗较低,能够提供更高的隐私保护程度。
针对联邦学习存在的隐私安全问题和收敛困难问题,提出了一种自适应差分隐私机制。经过仿真实验验证,得到如下结论。
(1)提出的方法可以根据不同的训练阶段自适应地调整学习率,提高了模型收敛效率,解决了联邦学习模型收敛困难的问题。
(2)引入了差分隐私技术,通过对训练梯度添加噪声有效保护了联邦学习的隐私安全。
(3)相较于其他先进方法,提出的自适应差分隐私机制可以在消耗较少的隐私预算的情况下,取得不错的准确率。
  • 内蒙古自治区高等学校科学研究项目(NJZY22382)
  • 内蒙古自治区直属高校基本科研业务费项目(JY20240010)
  • 内蒙古自治区直属高校基本科研业务费项目(JY20230082)
  • 内蒙古自治区自然科学基金(2023LHMS06016)
  • 内蒙古工业大学学科学研究项目(BS201936)
参考文献 引证文献
排序方式:
[1]
Mcmahan H B, Moore E, Ramage D, et al. Communication-efficient learning of deep networks from decentralized data[C]// Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. Fort Lauderdale: JMLR Workshop and Conference Proceedings, 2017: 1273-1282.
[2]
Dwork C. Differential privacy[C]//International Colloquium on automata, languages, and Programming. Berlin: Springer Berlin Heidelberg, 2006: 1-12.
[3]
Wei K, Li J, Ding M, et al. User-level privacy preserving federated learning: Analysis and performance optimization[J]. IEEE Transactions on Mobile Computing, 2021, 21(9): 3388-3401
[4]
M, Kim K. Differentially private federated learning via inexact admm[J]. arxiv preprint arxiv: 2106. 06127, 2021.
[5]
Huang X, Ding Y, Jiang Z L, et al. DP-FL: a novel differentially private federated learning framework for the unbalanced data[J]. World Wide Web, 2020, 23: 2529-2545.
[6]
Hu R, Guo Y, Ratazzi E P, et al. Differentially private federated learning for resource-constrained internet of things[J]. arXiv preprint arXiv: 2003. 12705, 2020.
[7]
Zhang J, Zhao Y, Wang J, et al. FedMEC: improving efficiency of differentially private federated learning via mobile edge computing[J]. Mobile Networks and Applications, 2020, 25: 2421-2433.
[8]
Leng J, Ye S, Zhou M, et al. Block chain-secured smart manufacturing in industry 4. 0: a survey[J]. IEEET ransactions on Systems, Man, and Cybernetics: Systems, 2020, 51(1): 237-252.
[9]
江欣俞, 李晓会, 秦若婷, . 基于图神经网络的兴趣点推荐的隐私保护框架[J]. 科学技术与工程, 2023, 23(17): 7407-7419.
Jiang Xinyu, Li Xiaohui, Qin Ruoting, et al. Privacy protection framework based on recommendation of points of interest[J]. Science Technology and Engineering, 2023, 23 (17): 7407-7419.
[10]
李洋, 徐进, 朱建明, . 可实现双向自适应差分隐私的联邦学习方案[J]. 西安电子科技大学学报, 2024, 51(3): 158-169.
Li Yang, Xu Jin, Zhu Jianming, et al. A federated learning scheme that can achieve bidirectional adaptive differential privacy[J]. Journal of Xidian University, 2024, 51 (3): 158-169.
[11]
Xu Z, Shi S, Liu A X, et al. An adaptive and fast convergent approach to differentially private deep learning[C]// IEEE INFOCOM 2020-IEEE Conference on Computer Communications. New York: IEEE, 2020: 1867-1876.
[12]
Ye D, Yu R, Pan M, et al. Federated learning in vehicular edge computing: a selective model aggregateon approach[J]. IEEE Access, 2020, 8: 23920-23935.
[13]
Liu Y, Kang Y, Xing C, et al. A secure federated transfer learning framework[J]. IEEE Intelligent Systems, 2020, 35(4): 70-82.
[14]
Dwork C, Roth A. The algorithmic foundations of differential privacy[J]. Foundations and Trends© in Theoretical Computer Science, 2014, 9(3/4): 211-407.
[15]
Yang Q, Liu Y, Chen T, et al. Federated machine learning: concept and applications[J]. ACM Transactions on Intelligent systems and Technology, 2019, 10(2): 1-19.
[16]
Mercier Q, Poirion F, Désidéri J A. A stochastic multiple gradient descent algorithm[J]. European Journal of Operational Research, 2018, 271(3): 808-817.
[17]
Kingma D P. Adam: A method for stochastic optimization[J]. arxiv preprint arxiv: 1412. 6980, 2014.
[18]
Fang W, Yao X, Zhao X, et al. A stochastic control-approach to maximize profit on service provisioning for mobile cloudlet platforms[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2016, 48(4): 522-534.
[19]
Ward R, Wu X, Bottou L. Adagrad stepsizes: Sharp convergence over nonconvex landscapes[J]. The Journal of Machine Learning Research, 2020, 21(1): 9047-9076.
[20]
Kurbiel T, Khaleghian S. Training of deep neural networks based on distance measures using RMSProp[J]. arXiv preprint arXiv: 1708. 01911, 2017.
[21]
Ding J, Ren X, Luo R, et al. An adaptive and momental bound method for stochastic learning[J]. arXiv preprint arXiv: 1910. 12249, 2019.
[22]
Abadi M, Chu A, Goodfellow I, et al. Deep learningwith differential privacy[C]// Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security. New York: ACM, 2016: 308-318.
2025年第25卷第7期
PDF下载
171
75
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2308502
  • 接收时间:2023-10-31
  • 首发时间:2026-03-30
  • 出版时间:2025-03-08
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2023-10-31
  • 修回日期:2024-07-26
基金
内蒙古自治区高等学校科学研究项目(NJZY22382)
内蒙古自治区直属高校基本科研业务费项目(JY20240010)
内蒙古自治区直属高校基本科研业务费项目(JY20230082)
内蒙古自治区自然科学基金(2023LHMS06016)
内蒙古工业大学学科学研究项目(BS201936)
作者信息
    内蒙古工业大学信息工程学院, 呼和浩特 010080

通讯作者:

* 石宝(1982—),男,蒙古族,内蒙古兴安盟人,博士,教授。研究方向:人工智能。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2308502
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏