The paper presents a framework for resource allocation in a mobile edge computing (MEC) system to provide customized virtual reality (VR) services to heterogeneous users. It introduces a quality of experience (QoE) metric to measure user experience considering system latency, user attention levels, and preferred resolutions. The resource allocation problem is formulated as a reinforcement learning problem to learn a generalized policy for all MEC servers across diverse user environments. The proposed framework employs federated learning and prompt-based sequence modeling to pre-train a decision model across MEC servers, named FedPromptDT. This approach solves local MEC data insufficiency and protects user privacy during offline training. The model’s adaptability to various user environments is improved by integrating user-environment cues and user-preferred allocation in the design of prompts. Experimental evaluations demonstrate that FedPromptDT outperforms baseline methods and exhibits remarkable adaptability across various user environments.
Publication date: 16 Feb 2024
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2402.09729
