Precio: Private Aggregate Measurement via Oblivious Shuffling

  • Melissa Chase, Microsoft

This talk presents Precio, a secure aggregation system for computing layered histograms and sums over secret shared data in a client-server setting. Precio is motivated by private ad measurement scenarios, where online advertisers and ad networks want to measure the performance of ad campaigns without requiring privacy-invasive techniques (third-party cookies). Technical conversations regarding mechanisms for private ad measurement continue actively in W3C and IETF; Precio provides one such mechanism, but we believe it can also be valuable in other contexts.

Precio has linear (time and communication) complexity in the number of data points and guarantees differentially private outputs. Compared to prior work it supports much larger domains and/or much more flexible aggregates, with up to four orders of magnitude better performance in some settings. A complete implementation is available at https://GitHub.com/Microsoft/Precio (opens in new tab).

In this talk we introduce the problem of private ad measurement, explain the approach that Precio takes and how it compares to prior proposals. Finally, we discuss open problems in this space.