Federated learning eliminates the necessity of transferring private training data and instead relies on the aggregation of model updates. Several publications on privacy attacks show how these individual model updates are vulnerable to the extraction of sensitive information. State-of-the-art secure aggregation protocols provide privacy for participating clients, yet, they are restrained by high computation and communication overhead.

We introduce S-BDT: a novel (𝜀, 𝛿)-differentially private distributed gradient boosted decision tree (GBDT) learner that improves the protection of single training data points (privacy) while achieving meaningful learning goals, such as accuracy or regression error (utility).S-BDT uses less noise by relying on non-spherical multivariate Gaussian noise, for which we show tight subsampling bounds for privacy amplification and incorporate that into a Rényi filter for individual privacy accounting.We experimentally reach the same utility while saving 50% in terms of epsilon for 𝜀 ≤ 0.5 on the Abalone regression dataset (dataset size ≈ 4𝐾), saving 30% in terms of epsilon for 𝜀 ≤ 0.08 for the Adult classification dataset (dataset size ≈ 50𝐾), and saving 30% in terms of epsilon for 𝜀 ≤ 0.03 for the Spambase classification dataset (dataset size ≈ 5𝐾).Moreover, we show that for situations where a GBDT is learning a stream of data that originates from different subpopulations (non-IID), S-BDT improves the saving of epsilon even further. CCS Concepts• Theory of computation → Theory of database privacy and security; • Computing methodologies → Boosting; Online learning settings.

The Automatic Dependant Surveillance-Broadcast (ADS-B) message scheme was designed without any authentication or encryption of messages in place. It is therefore easily possible to attack it, e.g., by injecting spoofed messages or modifying the transmitted Global Navigation Satellite System (GNSS) coordinates. In order to verify the integrity of the received information, various methods have been suggested, such as multilateration, the use of Kalman filters, group certification, and many others. However, solutions based on modifications of the standard may be difficult and too slow to be implemented due to legal and regulatory issues. A vantage far less explored is the location verification using public sensor data. In this paper, we propose LoVe, a lightweight message verification approach that uses a geospatial indexing scheme to evaluate the trustworthiness of publicly deployed sensors and the ADS-B messages they receive. With LoVe, new messages can be evaluated with respect to the plausibility of their reported coordinates in a location privacy-preserving manner, while using a datadriven and lightweight approach. By testing our approach on two open datasets, we show that LoVe achieves very low false positive rates (between 0 and 0.001 06) and very low false negative rates (between 0.000 65 and 0.003 34) while providing a real-time compatible approach that scales well even with a large sensor set. Compared to currently existing approaches, LoVe neither requires a large number of sensors, nor for messages to be recorded by as many sensors as possible simultaneously in order to verify location claims. Furthermore, it can be directly applied to currently deployed systems thus being backward compatible.