CIF: Small: Resource-Constrained Distributed Multiple Testing with False Discovery Rate Control

Distributed intelligence with large amounts of local measurement data hinges on the design of devices that are capable of sensing, processing, and exchanging local information. Reliable decision-making in a collective manner is, however, a challenging task in resource-constrained scenarios, where devices are limited by communication costs and/or processing power.

This project aims to develop a new framework for designing efficient decentralized algorithms such that the average proportion of wrong decisions in the network is bounded by a prescribed target threshold. This line of research has implications for a broad range of real-world applications, including environmental monitoring using battery-powered mobile sensors, coordination of unmanned aerial vehicles for target tracking, and multimedia wireless sensor networks in surveillance. The project will also provide mentoring and training of future algorithm designers.

This project investigates the structural properties of optimal decision rules under the false discovery rate (FDR) control, which provides guidance for new co-design of summary statistics and aggregation mechanisms, thereby enabling efficient decentralized processing in resource-constrained environments. The research program will explore three main thrusts: (i) develop communication-efficient algorithms (measured in bits) for multi-hop networks with provable FDR control; (ii) characterize computation-efficient approximations of the optimal decision rule in both the finite-sample and asymptotic regimes; and (iii) develop distributed feature selection with FDR control when all the features are shared among devices, focusing on privacy, robustness, and computation efficiency.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.