🤖 AI Summary
Metaverse-enabled wireless systems face core challenges in jointly optimizing Quality of Service (QoS) and Quality of Physical Experience (QoPE), including modeling complexity, high communication overhead, and stringent privacy constraints. To address these, this paper proposes Distributed Split Federated Learning (DS-FL), a novel paradigm that systematically characterizes the key machine learning requirements for metaverse wireless scenarios—namely, dynamic digital twin modeling, virtual avatar behavioral learning, and immersive interaction optimization. DS-FL achieves efficient, privacy-preserving model co-training via hierarchical model splitting, edge-cloud collaborative training, and channel-aware gradient compression. Simulation results demonstrate that DS-FL maintains over 92% convergence accuracy while reducing communication load by 40% compared to baseline methods, significantly enhancing training scalability and real-time performance. The framework establishes a deployable intelligent modeling infrastructure for metaverse wireless systems.
📝 Abstract
—Today’s wireless systems are posing key challenges in terms of quality of service and quality of physical experience. Metaverse has the potential to reshape, transform, and add inno- vations to the existing wireless systems. A metaverse is a collective virtual open space that can enable wireless systems using digital twins, digital avatars, and interactive experience technologies. Machine learning (ML) is indispensable for modeling twins, avatars, and deploying interactive experience technologies. In this paper, we present the role of ML in enabling metaverse- based wireless systems. We identify and discuss a set of key requirements for advancing ML in the metaverse-based wireless systems. Moreover, we present a case study of distributed split federated learning for efficiently training meta-space models. Finally, we discuss the future challenges along with potential solutions.
