Lead PI: Basak Guler

This thrust will focus on developing methods for deploying machine learning models in the field to support both network management and the early, preemptive detection of threats. A key emphasis is on ensuring that these models remain lightweight and can be retrained or fine-tuned during deployment with minimal communication and computational overhead. The effort is closely integrated with the goals of the other thrusts, particularly in enabling adaptive network operation and secure, resilient deployments in dynamic environments. 

To accomplish this, the thrust will develop machine learning models that support network control functions and resource allocation across the system. These models will be designed to operate at multiple levels, with fine-grained decision-making at edge nodes and broader resource coordination at more capable nodes, including the use of reinforcement learning methods tailored to different operational scales. At the same time, the thrust will create distributed network management mechanisms that allow models to be updated efficiently without overwhelming network resources, prioritizing urgent updates for near-real-time deployment while enabling routine refinements with low overhead. In parallel, it will advance distributed learning approaches for secure networking that can detect and respond to evolving cyber threats across heterogeneous environments, providing scalable and accurate methods for decentralized intrusion detection and rapid online model adaptation.