Abigail Gentle

Research

Lightweight Protocols for Distributed Private Quantile Estimation #
Anders Aamand, Fabrizio Boninsegna, Abigail Gentle, Jacob Imola, Rasmus Pagh
Adaptivity helps when algorithms operate under information constraints such as local differential privacy. For the quantiles problem (e.g. finding the median) this adaptivity is crucial to overcome lower bounds against CDF learning. Analysing this algorithm on a fixed population visited in random order requires a novel bound on the maximum CDF change. An algorithm in the shuffle model shows what can be achieved in fewer rounds of adaptivity.
Locally Private Histograms in All Privacy Regimes #
Clément Canonne, Abigail Gentle
Improved analysis of existing algorithms for locally private histograms shows optimal error in low-privacy regimes, implying optimal high-privacy algorithms in the shuffle model of differential privacy.