Smart cities promise safer streets, smoother traffic, and more efficient services, enabled by dense networks of urban sensors. Yet this infrastructure, often unnoticed by citizens, introduces pervasive privacy risks, from tracking, profiling, and sensitive inferences to subtle forms of self-censorship. Despite widespread deployment, little is known about how the public understands and perceives these sensing systems. To address this gap, we present an intervention based user study (n = 172) in which participants are exposed to data collection by six urban sensors, including cameras and alternative technologies commonly framed as privacy-preserving. Participants encounter either the sensors alone or sensors accompanied by real-time data visualizations. Our results reveal widespread misunderstanding of some sensors (radar, LiDAR, Wi-Fi, depth, and thermal imaging sensors), particularly their capacity for identification and for attribute inferences such as gender or age. We also identify persistent misconceptions, including the belief that Wi-Fi poses privacy risks only when users connect to public networks. While making sensors visible and visualizing collected data improves privacy awareness, these measures alone are not enough for citizens to understand the actual risks of urban sensing. We derive recommendations for privacy-respecting smart city environments grounded in citizens’ informational needs and expectations.

Beamforming, as introduced in WiFi 5, requires clients to broadcast observations of their channel characteristics. This introduces a new information source for WiFi sensing with privacy threats that have not been explored, so far. With WiFi networks being ubiquitous in our everyday lives, the impact of unknown privacy threats is likely severe. To investigate this concern, we introduce BFId, the first identity inference attack using BFI-based sensing and evaluate its efficacy on a novel dataset containing WiFi recordings of 197 individuals. We show that we can infer the identity of individuals with very high accuracy, across different walking styles and perspectives, even with large sample sizes.

Programmatic hyperparameter optimization (HPO) methods, such as Bayesian optimization and evolutionary algorithms, are known for their sample efficiency in identifying optimal configurations for machine learning (ML) models. However, practitioners often use less efficient methods, such as grid search, potentially resulting in under-optimized models. This discrepancy suggests that HPO method selection may be influenced by practitioner-specific motives, which remain insufficiently understood hindering user-centered advancement of HPO tools. To uncover these motives, we conducted 20 semi-structured interviews and an online survey with 49 ML practitioners. We revealed six primary goals (e.g., increasing ML model understanding) and 14 contextual factors (e.g., available computational resources) that influence practitioners’ choices of HPO methods. This study provides a conceptual foundation for understanding real-world HPO practices and informs the development of more user-centered and context-adaptive HPO tools in automated ML (AutoML).

Research on the effects of essential hyperparameters of DP-SGD lacks consensus, verification, and replication. Contradictory and anecdotal statements on their influence make matters worse. While DP-SGD is the standard optimization algorithm for privacy-preserving machine learning, its adoption is still commonly challenged by low performance compared to non-private learning approaches. As proper hyperparameter settings can improve the privacy-utility trade-off, understanding the influence of the hyperparameters promises to simplify their optimization towards better performance, and likely foster acceptance of private learning.To shed more light on these influences, we conduct a replication study: We synthesize extant research on hyperparameter influences of DP-SGD into conjectures, conduct a dedicated factorial study to independently identify hyperparameter effects, and assess which conjectures can be replicated across multiple datasets, model architectures, and differential privacy budgets. While we cannot (consistently) replicate conjectures about the main and interaction effects of the batch size and the number of epochs, we were able to replicate the conjectured relationship between the clipping threshold and learning rate. Furthermore, we were able to quantify the significant importance of their combination compared to the other hyperparameters.