Efficient and accurate strategies for differentially-private sliding window queries

Abstract

Regularly releasing the aggregate statistics about data streams in a privacy-preserving way not only serves valuable commercial and social purposes, but also protects the privacy of individuals. This problem has already been studied under differential privacy, but only for the case of a single continuous query that covers the entire time span, e.g., counting the number of tuples seen so far in the stream. However, most real-world applications are window-based, that is, they are interested in the statistical information about streaming data within a window, instead of the whole unbound stream. Furthermore, a Data Stream Management System (DSMS) may need to answer numerous correlated aggregated queries simultaneously, rather than a single one. To cope with these requirements, we study how to release differentially private answers for a set of sliding window aggregate queries. We propose two solutions, each consisting of query sampling and composition. We first selectively sample a subset of representative sliding window queries from the set of all the submitted ones. The representative queries are answered by adding Laplace noises in a way satisfying differential privacy. For each non-representative query, we compose its answer from the query results of those representatives. The experimental evaluation shows that our solutions are efficient and effective.

Differentially Private MIMO Filtering for Event Streams

“…5 are public information representing known aggregate statistics about the server dynamics. The numbering of states 1 to 4 is the one used to express the z-spectrum (25). We assume α, β / ∈ {0, 1}, in which case the chain is ergodic.…”

“…An application example is that of analyzing spatio-temporal records provided by networks of simple counting sensors, e.g., motion detectors in buildings or inductive-loop detectors in traffic information systems [24]. The literature on the differentially private processing of multidimensional time series is still very limited, but includes [25], which considers a single-input multiple-output filter where each output channel corresponds to a moving average filter with a different size for the averaging window, as well as [26], which discusses an application to traffic monitoring.…”

Differentially private event sequences over infinite streams

“…Cao et al [5] specify a set of range queries on the time domain before the system starts. Each query requests the sum of the updates therein.…”

“…The works that aim at event-level privacy in infinite streams [11,27,6,7] capture counter-based scenarios, whereas in our model each event contributes to a single published statistic. Cao et al [5] exploit overlapping query ranges, whereas we publish non-overlapping counts at each timestamp. Furthermore, all the schemes in the static scenario (described in Section 2.1) require a priori knowledge of all data, whereas we process the data on-the-fly as they arrive from the stream.…”

“…For any window of size 3, the sum of the budgets therein is at most ϵ. Note that, if BA does not nullify the budget at timestamp 4, the sum of the budgets in window [3,5] is 7ϵ/6 > ϵ, which violates w-event privacy.…”