KIT | KIT-Bibliothek | Impressum | Datenschutz

Swellfish Privacy: Exploiting Time-Dependent Relevance for Continuous Differential Privacy : Technical Report

Tex, Christine ORCID iD icon; Schäler, Martin; Böhm, Klemens

Abstract:

Today, continuous publishing of differentially private query results is the de-facto standard. The challenge hereby is adding enough noise to satisfy a given privacy level, and adding as little noise as necessary to keep high data utility. In this context, we observe that privacy goals of individuals vary significantly over time. For instance, one might aim to hide whether one is on vacation only during school holidays. This observation, named time-dependent relevance, implies two effects which – properly exploited – allow to tune data utility. The effects are time-variant sensitivity (TEAS) and time-variant number of affected query results (TINAR). As today’s DP frameworks, by design, cannot exploit these effects, we propose Swellfish privacy. There, with policy collections, individuals can specify combinations of time-dependent privacy goals. Then, query results are Swellfish-private, if the streams are indistinguishable with respect to such a collection.We propose two tools for designing Swellfish-private mechanisms, namely, temporal sensitivity and a composition theorem, each allowing to exploit one of the effects. In a realistic case study, we show empirically that exploiting both effects improves data utility by one to three orders of magnitude compared to state-of-the-art w-event DP mechanisms. ... mehr


Volltext §
DOI: 10.5445/IR/1000123707
Veröffentlicht am 16.09.2020
Cover der Publikation
Zugehörige Institution(en) am KIT Institut für Programmstrukturen und Datenorganisation (IPD)
Publikationstyp Forschungsbericht/Preprint
Publikationsmonat/-jahr 09.2020
Sprache Englisch
Identifikator ISSN: 2194-1629
KITopen-ID: 1000123707
Verlag Karlsruher Institut für Technologie (KIT)
Umfang 14 S.
Serie KIT Scientific Working Papers ; 151
Schlagwörter differential privacy, streams, monitoring, privacy policies, data utility, time-dependent relevance
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page