NSF-BSF:CIF: Localization Theories for Modern Quickest Detection

Quickest detection, also known as sequential change-point detection, has many important real-world applications, including biosurveillance, environmental monitoring, disaster response, homeland security, manufacturing quality control, and threat detection. The goal of quickest detection is to detect change in the area (e.g., a sensor) as quickly as possible, subject to a global constraint on the rate of false alarms. While the classical quickest detection theory is well developed for systems involving a single sensor or a small number of sensors, the fundamental information bounds and optimal design principles for modern data-rich environments remain largely unknown. This project will develop a general theoretical framework to characterize the information-theoretic limits of quickest detection and develop efficient online algorithms for modern settings involving high-dimensional data, distributed sensor networks, and/or privacy constraints. The results will provide rigorous foundations for future sensing and monitoring systems, and help practitioners assess when improvements in data quality or sensing techniques are necessary to meet application-specific requirements. Graduate students will also receive training through their involvement in the project's research. This project develops general localization theories for establishing information-theoretic bounds and statistical performance of quickest detection algorithms under modern asymptotic regimes and complex data structures. The central idea is a divide-and-conquer approach that partitions a global detection problem into a sequence of local sub-problems that can be investigated using localization theorems. The specific research aims are: (1) Information bounds on sparse change detection; (2) Federated quickest detection over networks; and (3) Quickest detection with differential privacy. In addition, real-world case studies in engineering and health will be investigated. Results of the project are expected to advance the state of the art in sequential analysis, change-point detection, streaming data analysis, online learning, and large-scale statistical inference. 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.