Distributed Autonomous Online Learning.pdf

Distributed Autonomous Online Learning.pdf

ID:33945283

大小:311.42 KB

页数:11页

时间:2019-03-01

Distributed Autonomous Online Learning.pdf_第1页
Distributed Autonomous Online Learning.pdf_第2页
Distributed Autonomous Online Learning.pdf_第3页
Distributed Autonomous Online Learning.pdf_第4页
Distributed Autonomous Online Learning.pdf_第5页
资源描述:

《Distributed Autonomous Online Learning.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、IEEETRANSACTIONSONKNOWLEDGEANDDATAENGINEERING,VOL.25,NO.11,NOVEMBER20132483DistributedAutonomousOnlineLearning:RegretsandIntrinsicPrivacy-PreservingPropertiesFengYan,ShreyasSundaram,Member,IEEE,S.V.N.Vishwanathan,andYuanQiAbstract—Onlinelearninghasbecomeincreasinglypopularonhandlingmassivedata.Th

2、esequentialnatureofonlinelearning,however,requiresacentralizedlearnertostoredataandupdateparameters.Inthispaper,weconsideronlinelearningwithdistributeddatasources.Theautonomouslearnersupdatelocalparametersbasedonlocaldatasourcesandperiodicallyexchangeinformationwithasmallsubsetofneighborsinacommu

3、nicationnetwork.WederivetheregretboundforstronglyconvexfunctionsthatgeneralizestheworkbyRametal.[1]forconvexfunctions.Moreimportantly,weshowthatouralgorithmhasintrinsicprivacy-preservingproperties,andweprovethesufficientandnecessaryconditionsforprivacypreservationinthenetwork.Theseconditionsimply

4、thatfornetworkswithgreater-than-oneconnectivity,amaliciouslearnercannotreconstructthesubgradients(andsensitiverawdata)ofotherlearners,whichmakesouralgorithmappealinginprivacy-sensitiveapplications.IndexTerms—Onlinelearning,distributedcomputing,privacypreservationÇ1INTRODUCTIONNLINElearninghasemer

5、gedasanattractiveparadigmpassestimationinformationtotheirneighborsinacom-Oinmachinelearninggiventheever-increasingmunicationnetwork.Bydoingso,distributedlearningamountsofdatabeingcollectedeveryday.Itefficientlyavoidssharingoriginal,sensitivedatawithothersandreducesthetrainingtimebyprocessingtheda

6、taonlyonce,storingdatainacentrallocation.assumingthatallthetrainingdataareavailableatacentralInthispaper,weconsiderageneraldistributedautono-location.Formanyapplications,however,thisassumptionmousonlinelearningalgorithmtolearnfromfullydecentra-isproblematic.Forinstance,sensornetworksmaybelizeddat

7、asources.Weaddresstwoimportantquestionsdeployedinrainforestsandcollectdataautonomously.associatedwiththisgeneralalgorithm.ThefirstquestionisThecostoftransmittingallthedatatoacentralservercanhowthedistributedonlinelearn

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。