Differential Privacy Budget Calculator

Differential Privacy Budget Calculator

animation
location
privacy
Try Change Privacy budget per query (ε): from 0.1 to 1.0.
Observe Watch Noise Standard Deviation and the 95% noise range around assumed true count.
Explain Explain why reducing ε from 1.0 increases Noise Standard Deviation.

Technical boundaries. Illustration only: count sensitivity is fixed at 1 and composition adds ε values. Composition and release risk also depend on the neighboring relation, sensitivity, implementation and context.

TipPutting Numbers to It

Context: Quantifying location privacy risks and protections.

Differential Privacy Budget Calculator

Explore the privacy/utility trade-off in differential privacy systems.

Key Understanding: Differential privacy adds calibrated noise under a stated mechanism and neighboring relation. For fixed sensitivity, a smaller ε means more Laplace noise and a stronger formal privacy bound; composition across queries adds the ε values.


Re-identification Attack Success Rate: Bike-sharing dataset: 500,000 trips with start/end stations and times (no user IDs). Researchers correlate: morning trips from residential areas + evening return trips + cross-reference with public voter registration. Worked example: 82% of regular commuters (trips >10 times) successfully re-identified. Attack works because spatial routine is quasi-identifier: home station narrows to ~1,000 residents, work station to ~200 employees, combining both to ~5-20 individuals. With 10 trips, temporal patterns (leaves 8:05am ± 5min daily) further narrow to 1-2 unique individuals. Lesson: removing user IDs insufficient—location trails themselves are identifiers.

Back to the chapter