Privacy-preserving statistical analysis using secure multi-party computation
Failid
Kuupäev
2015-02-06
Autorid
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Abstrakt
Kaasaegses ühiskonnas luuakse inimese kohta digitaalne kirje kohe pärast tema sündi. Sellest hetkest alates jälgitakse tema käitumist ning kogutakse andmeid erinevate eluvaldkondade kohta. Kui kasutate poes kliendikaarti, käite arsti juures, täidate maksudeklaratsiooni või liigute lihtsalt ringi mobiiltelefoni taskus kandes, koguvad ning salvestavad firmad ja riigiasutused teie tundlikke
andmeid.
Vahel anname selliseks jälitustegevuseks vabatahtlikult loa, et saada mingit kasu. Näiteks võime saada soodustust, kui kasutame kliendikaarti. Teinekord on meil vaja teha keeruline otsus, kas loobuda võimalusest teha mobiiltelefonikõnesid
või lubada enda jälgimine mobiilimastide kaudu edastatava info abil. Riigiasutused haldavad infot meie tervise, hariduse ja sissetulekute kohta, et meid paremini ravida, harida ja meilt makse koguda. Me loodame, et meie andmeid kasutatakse mõistlikult, aga samas eeldame, et meie privaatsus on tagatud.
Käesolev töö uurib, kuidas teostada statistilist analüüsi nii, et tagada üksikisiku
privaatsus. Selle eesmärgi saavutamiseks kasutame turvalist ühisarvutust. See krüptograafiline meetod lubab analüüsida andmeid nii, et üksikuid väärtuseid ei ole kunagi võimalik näha. Hoolimata sellest, et turvalise ühisarvutuse kasutamine on aeganõudev protsess, näitame, et see on piisavalt kiire ja seda on võimalik kasutada isegi väga suurte andmemahtude puhul.
Me oleme teinud võimalikuks populaarseimate statistilise analüüsi meetodite kasutamise turvalise ühisarvutuse kontekstis. Me tutvustame privaatsust säilitavat statistilise analüüsi tööriista Rmind, mis sisaldab kõiki töö käigus loodud funktsioone. Rmind sarnaneb tööriistadele, millega statistikud on harjunud. See lubab neil viia läbi uuringuid ilma, et nad peaksid üksikasjalikult tundma allolevaid krüptograafilisi protokolle.
Kasutame dissertatsioonis kirjeldatud meetodeid, et valmistada ette statistiline
uuring, mis ühendab kaht Eesti riiklikku andmekogu. Uuringu eesmärk on teada saada, kas Eesti tudengid, kes töötavad ülikooliõpingute ajal, lõpetavad nominaalajaga väiksema tõenäosusega kui nende õpingutele keskenduvad kaaslased.
In a modern society, from the moment a person is born, a digital record is created. From there on, the person’s behaviour is constantly tracked and data are collected about the different aspects of his or her life. Whether one is swiping a customer loyalty card in a store, going to the doctor, doing taxes or simply moving around with a mobile phone in one’s pocket, sensitive data are being gathered and stored by governments and companies. Sometimes, we give our permission for this kind of surveillance for some benefit. For instance, we could get a discount using a customer loyalty card. Other times we have a difficult choice – either we cannot make phone calls or our movements are tracked based on cellular data. The government tracks information about our health, education and income to cure us, educate us and collect taxes. We hope that the data are used in a meaningful way, however, we also have an expectation of privacy. This work focuses on how to perform statistical analyses in a way that preserves the privacy of the individual. To achieve this goal, we use secure multi-‐party computation. This cryptographic technique allows data to be analysed without seeing the individual values. Even though using secure multi-‐party computation is a time-‐consuming process, we show that it is feasible even for large-‐scale databases. We have developed ways for using the most popular statistical analysis methods with secure multi-‐party computation. We introduce a privacy-‐preserving statistical analysis tool called Rmind that contains all of our resulting implementations. Rmind is similar to tools that statistical analysts are used to. This allows them to carry out studies on the data without having to know the details of the underlying cryptographic protocols. The methods described in the thesis are used in practice to prepare for running a statistical study on large-‐scale real-‐life data to find out whether Estonian students who are working during university studies are less likely to graduate in nominal time.
In a modern society, from the moment a person is born, a digital record is created. From there on, the person’s behaviour is constantly tracked and data are collected about the different aspects of his or her life. Whether one is swiping a customer loyalty card in a store, going to the doctor, doing taxes or simply moving around with a mobile phone in one’s pocket, sensitive data are being gathered and stored by governments and companies. Sometimes, we give our permission for this kind of surveillance for some benefit. For instance, we could get a discount using a customer loyalty card. Other times we have a difficult choice – either we cannot make phone calls or our movements are tracked based on cellular data. The government tracks information about our health, education and income to cure us, educate us and collect taxes. We hope that the data are used in a meaningful way, however, we also have an expectation of privacy. This work focuses on how to perform statistical analyses in a way that preserves the privacy of the individual. To achieve this goal, we use secure multi-‐party computation. This cryptographic technique allows data to be analysed without seeing the individual values. Even though using secure multi-‐party computation is a time-‐consuming process, we show that it is feasible even for large-‐scale databases. We have developed ways for using the most popular statistical analysis methods with secure multi-‐party computation. We introduce a privacy-‐preserving statistical analysis tool called Rmind that contains all of our resulting implementations. Rmind is similar to tools that statistical analysts are used to. This allows them to carry out studies on the data without having to know the details of the underlying cryptographic protocols. The methods described in the thesis are used in practice to prepare for running a statistical study on large-‐scale real-‐life data to find out whether Estonian students who are working during university studies are less likely to graduate in nominal time.
Kirjeldus
Väitekirja elektrooniline versioon ei sisalda publikatsioone.
Märksõnad
statistiline analüüs, konfidentsiaalne info, privaatsus, Sharemind (tarkvara), statistical analysis, confidential information, privacy, Sharemind (software)