Vogue: Faster Computation of Private Heavy Hitters

Pranav Jangir, Nishat Koti, Varsha Bhat Kukkala, Arpita Patra, Bhavish Raj Gopal, Somya Sangal

Abstract: Consider the problem of securely identifying $\tau$-heavy hitters, where given a set of client inputs, the goal is to identify those inputs which are held by at least $\tau$ clients in a privacy-preserving manner. Towards this, we design a novel system $\mathsf {Vogue}$, whose key highlight in comparison to prior works, is that it ensures complete privacy and does not leak any information other than the heavy hitters. In doing so, $\mathsf {Vogue}$ aims to achieve as efficient a solution as possible. To showcase these efficiency improvements, we benchmark our solution and observe that it requires around 14 minutes to compute the heavy hitters for $\tau$ = 100 on 256-bit inputs when considering 400K clients. This is in contrast to the state of the art solution that requires over an hour for the same. In addition to the static input setting described above, $\mathsf {Vogue}$ also accounts for streaming inputs and provides a protocol that outperforms the state-of-the-art therein. The efficiency improvements witnessed while computing heavy hitters in both, the static and streaming input settings, are attributed to our new secure stable compaction protocol, whose round complexity is independent of the size of the input array to be compacted.