Calibrating Data to Sensitivity in Private Data Analysis: Davide Proserpio, BU

10:00 am on Monday, April 22, 2013
12:00 pm on Monday, April 22, 2013
MCS 137
Abstract:We present a new approach to differentially private computation in which one does not scale up the magnitude of noise for challenging queries, but rather scales down the contributions of challenging records. While scaling down all records uniformly is equivalent to scaling up the noise magnitude, we show that scaling records non-uniformly can result in substantially higher accuracy by bypassing the worst-case requirements of differential privacy for the noise magnitudes. In this paper we detail the data analysis platform wPINQ, which generalize the Privacy Integrated Query (PINQ) to weighted datasets. Using a few simple operators (including a non-uniformly scaling Join operator) we are able to reconstruct several recent results on graph analysis and introduce new generalizations,e.g. counting triangles with given degrees). We also show how to integrate probabilistic inference techniques to synthesize datasets respecting more complicated (and less easily interpreted) measurements, by connecting differential privacy and incremental re-computation. Joint work with Sharon Goldberg and Frank McSherry