On the Extremal Mechanisms for Local Differential Privacy & Binary Maximal Leakage
We consider two privacy measures, local differential privacy and binary maximal leakage, simultaneously. We study the subsequent utility-privacy trade-off for two utility problems: information preservation and f-divergence maximization. We present new achievability results for both problems over all mechanisms that satisfy the combined privacy measure and compare this with the optimal utility over all ϵ-LDP mechanisms. In particular, we study the binary mechanism with erasure and the randomized response mechanism with erasure. Finally, we show that singular mechanisms are dominating for binary maximal leakage.
École Polytechnique Fédérale de Lausanne
École Polytechnique Fédérale de Lausanne
2025-06-22
979-8-3315-4399-0
REVIEWED
EPFL
| Event name | Event acronym | Event place | Event date |
ISIT 2025 | Ann Arbor, MI, USA | 2025-06-22 - 2025-06-27 | |