大数据时代的隐私保护的挑战和机遇.pdf

《大数据时代的隐私保护的挑战和机遇.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、大数据时代的隐私保护：挑战和机遇张振杰博士伊利诺伊大学高等数字科学中心Outline•Whyprivacy?•PrivacyAttackingExamples•Oldmodelsandlimitations–K-anonymity–L-diversity•DifferentialPrivacyPublishingsensitivedataaboutindividuals.•Medicalresearch–Whattreatmentshavethebestoutcomes?–Howcanwerecognizetheonsetofdiseaseearlier?–Arecerta

2、indrugsbetterforcertainphenotypes?•Websearch–Whatarepeoplereallylookingforwhentheysearch?–Howcanwegivethemthemostauthoritativeanswers?•Publichealth–Whereareouroutbreaksofunpleasantdiseases?–Whatbehaviorpatternsorpatientcharacteristicsarecorrelatedwiththesediseases?Today,accesstothesedatase

3、tsisusuallystrictlycontrolled.Onlyavailable:•Insidethecompany/agencythatcollectedthedataSocietywould•Oraftersigningalegalcontractbenefitifwecould–Clickstreams,taxidata•Orinverycoarse-grainedsummariespublishsomeuseful–Publichealthformofthedata,•Orafteraverylongwaitwithouthavingto–USCensusda

4、tadetails•Orwithdefiniteprivacyissuesworryaboutprivacy.–USCensusreports,theAOLclickstream,oldNIHdbGaPsummarytables,Enronemail•OrwithIRB(InstitutionalReviewBoard)approval–dbGaPsummarytablesWhyisaccesssostrictlycontrolled?Nooneshouldlearnwhohadwhichdisease.“Microdata”Whatifwe“de-identify”the

5、recordsbyremovingnames?publishWecanre-identifypeople,absolutelyorprobabilisticallyAvoterregistrationlistThepublishedtableQuasi-identifier(QI)attributes“Backgroundknowledge”87%ofAmericanscanbeuniquelyidentifiedby{zipcode,gender,dateofactually63%birth}.[Golle06]LatanyaSweeney[InternationalJo

6、urnalonUncertainty,FuzzinessandKnowledge-basedSystems,2002]usedthisapproachtore-identifythemedicalrecordofanex-governorofMassachusetts.Outline•Whyprivacy?•PrivacyAttackingExamples•Oldmodelsandlimitations–K-anonymity–L-diversity•DifferentialPrivacyRealquerylogscanbeveryusefultoCSresearchers

7、.Butclickhistorycanuniquelyidentifyaperson.WhattheNewYorkTimesdid:–FindalllogentriesforAOLuser4417749–MultiplequeriesforbusinessesandservicesinLilburn,GA(population11K)–SeveralqueriesforJarrettArnold•Lilburnha