Differentially Private Release of Spatio-Temporal Data Statistics

# 217






Abstract

In this talk, we shall discuss some of the work done at India Urban Data Exchange on releasing statistics derived from spatio-temporal datasets in a user-level differentially private (DP) manner. Specifically, we will go over strategies for the user-level DP release of the sample mean and variance of a dataset, while keeping the error in reconstructing the statistic low. Of particular focus in the talk will be our novel, computable metric of the “worst-case error” in the reconstruction of a statistic, under a broad class of strategies. This, in turn, inspires a natural “error-preserving” algorithm for improving the privacy loss via the composition of user-level DP mechanisms acting on disjoint spatio-temporal regions, from a classical bound. We illustrate the efficacy of our mechanisms on real-world Intelligent Traffic Management System (ITMS) data from an Indian city.

Dr. V. Arvind Rameshwar, Research Fellow, Indian Urban Data Exchange

V. Arvind Rameshwar is a Research Fellow at India Urban Data Exchange, where he works on differential privacy. He received the B.E. (Hons.) degree in Electronics and Communication Engineering from BITS Pilani University, India, in 2018, and the Ph.D. degree from the Department of Electrical Communication Engineering, Indian Institute of Science, in 2023. During his graduate studies, he was a recipient of the Prime Minister's Research Fellowship 2020 and was part of teams that won Qualcomm Innovation Fellowships India 2020, 2022, and 2023. His papers have won the IEEE Jack Keil Wolf ISIT Student Paper Award and paper awards at the National Conference on Communications (NCC) and the IEEE International Conference on Signal Processing and Communications (SPCOM). His research interests lie broadly in information theory and error-control coding for non-standard channels and differential privacy.