Collaborative Research: Federated Bilevel Optimization over Wireless Networks

Emerging edge intelligence systems are transforming applications such as AI-native wireless networks, autonomous systems, smart manufacturing, and personalized healthcare by enabling distributed devices to collaboratively learn and make decisions. These systems rely on hierarchical decision-making over resource-constrained wireless networks, where global objectives depend on coordinated optimization performed locally across many edge devices. Federated bilevel optimization is an emerging mathematical framework for solving these hierarchical optimization problems while preserving data privacy. However, existing approaches are not designed for practical wireless environments, where unreliable communications, limited radio resources, constrained device capabilities, and dynamic operating conditions fundamentally affect optimization performance. This project develops the scientific foundations of communication-aware federated bilevel optimization over wireless networks to enable reliable, scalable, and adaptive edge intelligence. The project prepares the future workforce by integrating research outcomes into undergraduate and graduate courses at Washington State University and University of Houston, engaging undergraduate students in research, and conducting outreach activities that broaden participation in science, technology, engineering, and mathematics. The project also promotes interdisciplinary collaborations that evaluate the developed methods in industrial automation and large-scale scientific facilities, while releasing open-source software and educational resources to accelerate research and workforce development in intelligent wireless systems. This project aims to establish the foundations for federated bilevel optimization over wireless networks with resource-constrained devices, through a comprehensive investigation on performance characterization over noisy fading wireless channels and resource management. The research efforts are organized around two complimentary thrusts: (i) Thrust 1 investigates federated offline bilevel optimization with stationary functions over wireless networks, by characterizing the impact of communication and computation constraints on learning performance and designing adaptive resource management strategies to improve efficiency. The research also develops scalable semi-decentralized and hierarchical architectures to reduce communication overhead. (ii) Thrust 2 studies federated online bilevel optimization with nonstationary functions over wireless networks, by developing efficient optimization algorithms, adaptive client scheduling and resource allocation methods, and communication-efficient mechanisms for dynamic wireless environments. By bridging the fields of wireless networks and bilevel optimization, this project provides new frameworks for analyzing learning performance under communication/computation constraints, while advancing the understanding of resource management in hierarchical optimization structures. 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.