Practical locally private heavy hitters

Raef Bassily, Kobbi Nissim, Uri Stemmer, Abhradeep Thakurta

Research output: Contribution to journalArticlepeer-review

28 Scopus citations

Abstract

We present new practical local differentially private heavy hitters algorithms achieving optimal or near-optimal worst-case error and running time - TreeHist and Bitstogram. In both algorithms, server running time is Õ(n) and user running time is Õ(1), hence improving on the prior state-of-the-art result of Bassily and Smith [STOC 2015] requiring O(n5/2) server time and O(n3/2) user time. With a typically large number of participants in local algorithms (n in the millions), this reduction in time complexity, in particular at the user side, is crucial for making locally private heavy hitters algorithms usable in practice. We implemented Algorithm TreeHist to verify our theoretical analysis and compared its performance with the performance of Google's RAPPOR code.

Original languageEnglish
JournalJournal of Machine Learning Research
Volume21
StatePublished - 1 Feb 2020
Externally publishedYes

Funding

FundersFunder number
National Science FoundationAF-1908281, SHF-1907715
Google
Ohio State University1565387
Israel Science Foundation1871/19, TRIPODS-1740850

    Keywords

    • Differential privacy
    • Heavy hitters
    • Histograms
    • Local differential privacy
    • Sketching

    Fingerprint

    Dive into the research topics of 'Practical locally private heavy hitters'. Together they form a unique fingerprint.

    Cite this