Federated Prompt-Based Decision Transformer for Resource Allocation of Customized VR Streaming in Mobile Edge Computing

This paper investigates resource allocation for providing heterogeneous users with customized virtual reality (VR) streaming services in a mobile edge computing (MEC) system. We introduce a quality of experience (QoE) metric that considers system latency, user attention levels, and preferred resolutions to measure user experience based on the Weber-Fechner Law. A QoE maximization problem is then formulated for resource allocation to optimize user experience. It is cast as a reinforcement learning problem, aiming to learn a generalized policy applicable across diverse user environments of different MEC servers. To solve the problem, we propose a FedPromptDT framework, which employs federated learning (FL) and prompt-based generative sequence modeling to pre-train a common decision model across MEC servers. FL addresses the issue of insufficient local MEC data while protecting user privacy during offline training. Meanwhile, by integrating user-environment cues and user-preferred allocation, the design of prompts enhances the model’s adaptability to various user environments during online execution. Extensive experimental evaluations demonstrate that FedPromptDT outperforms baseline methods, exhibiting remarkable adaptability and maintaining superior performance across various user environments.