Alapfogalmak
Lower bounds for differential privacy under continual observation and online threshold queries are crucial for understanding the price of privacy over time.
Kivonat
The content discusses the private counter problem and its extension to the online thresholds problem. It presents new lower bounds and implications for various scenarios, including online learning and prediction models.
- Introduction to Differential Privacy and Continual Observation Model
- Differential privacy aims to protect individual-level information in statistical analysis.
- Continual observation model focuses on updating statistics while gathering sensitive data.
- Private Counter Problem
- Tracking events over time while maintaining privacy.
- Upper bounds and lower bounds on error dependence on time steps and events.
- Implications of Lower Bounds
- Extension to online thresholds problem and resolution of open questions.
- Separation between private and non-private online learners.
- Applications and Contributions
- Lower bounds for online counting and threshold monitor problems.
- Separation between private online learning and prediction models.
Statisztikák
Dwork et al. (2015) demonstrated an upper bound of O(log(T) + log2(n)).
Henzinger et al. (2023) showed a lower bound of Ω(min{log n, log T}).
Idézetek
"We show a new lower bound of Ω(min{n, log T}), which is tight w.r.t. the dependence on T."
"Our lower bound extends to the online thresholds problem, resolving an open question."