Conformalized Frequency Estimation from Sketched Data

Published in Advances in Neural Information Processing Systems 35 (NeurIPS), 2022

Abstract

A flexible conformal inference method is developed to construct confidence intervals for the frequencies of queried objects in a very large data set, based on the information contained in a much smaller sketch of those data. The approach is completely data-adaptive and makes no use of any knowledge of the population distribution or of the inner workings of the sketching algorithm; instead, it constructs provably valid frequentist confidence intervals under the sole assumption of data exchangeability. Although the proposed solution is much more broadly applicable, this paper explicitly demonstrates its use in combination with the famous count-min sketch algorithm and a non-linear variation thereof to facilitate the exposition. The performance is compared to that of existing frequentist and Bayesian alternatives through several experiments with synthetic data as well as with real data sets consisting of SARS-CoV-2 DNA sequences and classic English literature.

Download paper here