Edge Learning via Federated Split Decision Transformers for Metaverse Resource Allocation

TL;DR

This paper introduces FSDT, a federated split decision transformer framework that improves resource allocation in MEC-based metaverse services by reducing communication overhead and enhancing QoE under latency constraints.

Contribution

It proposes a novel federated split transformer model that partitions the transformer between MEC servers and the cloud, enabling local adaptability and cooperative training for resource allocation.

Findings

FSDT improves QoE by up to 10% in heterogeneous environments.

Nearly 98% of transformer parameters are offloaded to the cloud.

Reduces computational burden on MEC servers.

Abstract

Mobile edge computing (MEC) based wireless metaverse services offer an untethered, immersive experience to users, where the superior quality of experience (QoE) needs to be achieved under stringent latency constraints and visual quality demands. To achieve this, MEC-based intelligent resource allocation for virtual reality users needs to be supported by coordination across MEC servers to harness distributed data. Federated learning (FL) is a promising solution, and can be combined with reinforcement learning (RL) to develop generalized policies across MEC-servers. However, conventional FL incurs transmitting the full model parameters across the MEC-servers and the cloud, and suffer performance degradation due to naive global aggregation, especially in heterogeneous multi-radio access technology environments. To address these challenges, this paper proposes Federated Split Decision Transformer (FSDT), an offline RL framework where the transformer model is partitioned between MEC servers and the cloud. Agent-specific components (e.g., MEC-based embedding and prediction layers) enable local adaptability, while shared global layers in the cloud facilitate cooperative training across MEC servers. Experimental results demonstrate that FSDT enhances QoE for up to 10% in heterogeneous environments compared to baselines, while offloadingnearly 98% of the transformer model parameters to the cloud, thereby reducing the computational burden on MEC servers.