BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Audra McMillan (Apple) DTSTART:20210929T170000Z DTEND:20210929T180000Z DTSTAMP:20260423T021016Z UID:TCSPlus/29 DESCRIPTION:Title: Hiding among the clones: a simple and nearly optimal analysis of privacy amplification by shuffling\nby Audra McMillan (Apple) as part of TCS+\ n\n\nAbstract\nDifferential privacy (DP) is a model of privacy-preserving machine learning that has garnered significant interest in recent years du e to its rigorous privacy guarantees. An algorithm differentially private if the output is stable under small changes in the input database. While D P has been adopted in a variety of applications\, most applications of DP in industry actually satisfy a stronger notion called local differential p rivacy. In local differential privacy data subjects perturb their data bef ore it reaches the data analyst. While this requires less trust\, it comes a substantial cost to accuracy. Recent work of Erlingsson\, Feldman\, Mir onov\, Raghunathan\, Talwar\, and Thakurta [EFMRTT19] demonstrated that ra ndom shuffling amplifies differential privacy guarantees of locally random ized data. Such amplification implies substantially stronger privacy guara ntees for systems in which data is contributed anonymously [BEMMRLRKTS17] and has led to significant interest in the shuffle model of privacy [CSUZZ 19\, EFMRTT19]. In this talk\, we will discuss a new result on privacy amp lification by shuffling\, which achieves the asymptotically optimal depend ence in the local privacy parameter. Our result is based on a new proof st rategy which is simpler than previous approaches\, and extends to a lightl y weaker notion known as approximate differential privacy with nearly the same guarantees. \n\nBased on joint work with Vitaly Feldman and Kunal Tal war.\n LOCATION:https://researchseminars.org/talk/TCSPlus/29/ END:VEVENT END:VCALENDAR