Optimal Real-Time Fusion of Time-Series Data Under R\'enyi Differential Privacy

Researchers Chuanghong Weng and Ehsan Nekouei have published a significant paper on arXiv titled 'Optimal Real-Time Fusion of Time-Series Data Under Rényi Differential Privacy.' The work addresses a critical challenge in modern sensor networks and IoT systems: how to combine data from multiple correlated sensors in real-time while rigorously protecting individual privacy. The authors consider a setup where a fusion center receives private sensor measurements and must release aggregated outputs to an 'honest-but-curious' party trying to estimate an underlying state. Unlike previous approaches that apply privacy mechanisms statically, this research introduces a dynamic, optimization-based framework where privacy becomes a resource to be managed adaptively.

The core innovation is formulating privacy-aware fusion as a constrained finite-horizon optimization problem, jointly designing the fusion policy and state estimator to minimize estimation error subject to a total Rényi differential privacy budget. The derived optimal policy fundamentally differs from classical differential privacy by adaptively allocating privacy resources over time and operating in a closed-loop manner to regulate the adversary's beliefs. To make the solution computationally tractable, the researchers parameterized the fusion policy using a structured Gaussian distribution and developed a numerical optimization algorithm. They validated their framework through a traffic density estimation case study, showing practical effectiveness for applications ranging from smart transportation to industrial IoT, where real-time data fusion must balance utility with strong privacy guarantees.