Learning Based Resource Allocation
Learning-based resource allocation optimizes the distribution of limited resources (e.g., bandwidth, computing power) across competing demands, aiming to maximize overall system efficiency and performance. Current research heavily utilizes deep reinforcement learning (DRL), often incorporating graph neural networks (GNNs) to model complex dependencies between resources and users, and employing algorithms like soft actor-critic (SAC) or variations of Q-learning. This approach is proving valuable in diverse applications, from optimizing communication networks (e.g., 5G, IoV) and cloud computing to improving control allocation in aerospace systems, demonstrating significant improvements over traditional methods.
Papers
Zeroth-order Asynchronous Learning with Bounded Delays with a Use-case in Resource Allocation in Communication Networks
Pourya Behmandpoor, Marc Moonen, Panagiotis Patrinos
A Deep Learning Based Resource Allocator for Communication Systems with Dynamic User Utility Demands
Pourya Behmandpoor, Mark Eisen, Panagiotis Patrinos, Marc Moonen