Extending w-event Differential Privacy to Publish-Subscribe Systems

Sat, 22 Aug 2026 00:00:00 +0000

Internet of Things (IoT) devices have become ubiquitous in both enterprises and homes, and they increasingly rely on publish-subscribe (pub-sub) frameworks that give subscribers real-time, high-fidelity access to device data. For privacy and confidentiality, however, today’s pub-sub systems largely stop at channel-level authentication, encryption, and access-control lists. These protect data in transit and restrict who may receive it, but place no quantitative limit on what a subscriber with legitimate access to messages can infer about the publishers behind them.

This project closes that gap by extending w-event differential privacy to the pub-sub setting. We generalize w-event privacy from single-count streams to aggregate queries over a topic hierarchy, and introduce a population-aware budget-allocation strategy that calibrates noise from an operator-declared value range. Together these yield a provable w-event ε-DP guarantee that protects publisher identity across any window of w successive time instants, rather than a single event in isolation.

We realize the mechanism as a plugin to an MQTT broker, so the guarantee is enforced transparently such that existing publishers, subscribers, and brokers run unmodified. Evaluating on six real-world IoT datasets spanning energy, traffic, wearable health, air quality, mobility, and manufacturing, we show that the released error follows the predicted noise scale and that, at its tuned configuration, the mechanism delivers lower error than per-publisher local differential privacy while providing strong publisher privacy protections.

Differential Privacy | DIPr Lab at PSU

Extending w-event Differential Privacy to Publish-Subscribe Systems