This project is sponsored by the NSF under grant 2542859, “CAREER: Empowering a Resilient Edge Network with Proactive Traffic Management”.
Project Abstract
Data-driven applications increasingly shape how people interact with digital services, physical environments, and intelligent systems. Many of these applications rely on artificial intelligence (AI) to support real-time decision making, interactive experiences, and continuous sensing. These applications must operate alongside traditional Internet communication, including web access, cloud services, and background data transfer. To meet growing performance and responsiveness demands, both computation and networking are moving closer to users through edge networks. As a result, new AI-driven data flows coexist and compete with established network traffic on shared devices with limited resources. Current network management approaches struggle with this combination of heterogeneous workloads and multitasking devices. They also fail to account for changing user context and rely largely on reactive monitoring rather than anticipation. This project addresses these challenges by developing methods that enable edge networks to adapt proactively. The goal is to improve performance, efficiency, and reliability for AI-enabled services that support education, healthcare, transportation, immersive media, and other societal needs. Meanwhile, the project supports the development of Ph.D. and undergraduate students. Research outcomes are also integrated into academic courses and outreach activities.
The technical aim of the project is to develop and optimize an AI-native approach to edge network management. The research advances an intent-centric control framework in which the network learns context-aware representations of application behavior. These behaviors include inference tasks, collaborative model updates, sensing pipelines, and interactive workloads. Leveraging predictive and multi-timescale learning techniques, this project derives new methods to anticipate how application performance depends on time-varying interactions between communication and computation resources at the edge. This predictive capability enables adaptive orchestration across devices, edge servers, and cloud infrastructure. Building on general abstractions for coordinating heterogeneous AI workloads through learning-driven control and resource optimization, the research further develops methods that support hybrid task execution and collaborative learning while ensuring coexistence with traditional Internet traffic. All proposed methods are evaluated under realistic user activity conditions using representative workloads, including immersive media, digital twins, embodied systems, hybrid generative processing, and collaborative learning. Collectively, the results from this project establish a foundation for predictive, learning-based edge management.
Publications
- TBA
Acknowledgments
