Offline Diffusion Policy for Multi-User Delay-Constrained Scheduling

The article presents the Scheduling By Offline Learning with Critic Guidance and Diffusion Model (SOCD), an offline reinforcement learning algorithm designed for multi-user delay-constrained scheduling. SOCD utilizes a diffusion policy and a sampling-free critic network to derive efficient scheduling policies from pre-collected offline data, circumventing the need for online training which can degrade performance. Experimental results indicate that SOCD outperforms existing methods in various dynamic environments, making it a significant advancement for practitioners in resource allocation tasks within AI applications.