MetaFed: Advancing Privacy, Performance, and Sustainability in Federated Metaverse Systems
Muhammet Anil Yagiz 1 , Zeynep Sude Cengiz 1 , Polat Goktas 2 · 2025
Read MetaFed: Advancing Privacy, Performance, and Sustainability in Federated Metaverse Systems by Muhammet Anil Yagiz 1 , Zeynep Sude Cengiz 1 , Polat Goktas 2 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-17341.txt.
Introduction
Muhammet Anil Yagiz, Zeynep Sude Cengiz, and Polat Goktas present MetaFed in 2025, a federated learning framework for the Metaverse. Immersive virtual worlds strain centralized systems with lag, energy use, and privacy risk. Federated learning trains models on user devices instead of pooling raw data, and MetaFed adapts that idea to virtual world workloads.
Three parts work together: multi-agent reinforcement learning picks which clients train, encryption lets computation happen on protected data, and scheduling shifts work toward times with renewable energy. Tests on two image data sets with light networks cut carbon emissions by up to twenty-five percent versus standard approaches, while holding accuracy high and communication low. The design targets greener, private virtual infrastructure.
Catalog record
| CREATOR | Muhammet Anil Yagiz 1 , Zeynep Sude Cengiz 1 , Polat Goktas 2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17341
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MetaFed: Advancing Privacy, Performance, and Sustainability in Federated Metaverse Systems Muhammet Anil Yagiz 1 , Zeynep Sude Cengiz 1 , Polat Goktas 2 Abstract The rapid expansion of immersive Metaverse applications introduces complex challenges at the intersection of performance, privacy, and environmental sustainability. Centralized architectures fall short in addressing these demands, often resulting in elevated energy consumption, latency, and privacy concerns. This paper proposes MetaFed , a decentralized federated learning (FL) framework that enables sustainable and intelligent resource orchestration for Metaverse environments. MetaFed integrates (i) multi-agent reinforcement learning for dynamic client selection, (ii) privacy-preserving FL using homomorphic encryption, and (iii) carbon-aware scheduling aligned with renewable energy availability. Evaluations on MNIST and CIFAR-10 using lightweight ResNet architectures demonstrate that MetaFed achieves up to 25% reduction in carbon emissions compared to conventional approaches, while maintaining high accuracy and minimal communication overhead. These results highlight MetaFed as a scalable solution for building environmentally responsible and privacy-compliant Metaverse infrastructures. The source code and supplementary materials supporting this study are publicly accessible via our GitHub repository. 1 1 1 Available at: https://github.com/afrilab/MetaFed-FL I Introduction The Metaverse is emerging as a transformative paradigm, enabling immersive and persistent virtual experiences through an amalgamation of technologies such as extended reality (XR), blockchain, and digital twins [ 1 , 2 ] . These environments rely heavily on real-time rendering, multi-modal interaction, and high-fidelity simulation, often requiring up to 50 times more compute than traditional applications [ 3 ] . Supporting this scale of computation, especially across millions of users, necessitates robust infrastructures that are not only high-performing but also energy-efficient and privacy-conscious. Contemporary cloud-centric models suffer from significant drawbacks. First, the centralization of data and resources leads to severe scalability bottlenecks , as data centers struggle to manage latency-sensitive and compute-heavy Metaverse workloads [ 4 , 5 , 6 ] . Second, the aggregation of sensitive user data, including biometric, behavioral, and spatial information, triggers serious privacy concerns , especially in jurisdictions governed by strict data protection laws [ 7 ] . Third, the environmental impact of these infrastructures is non-trivial; data centers are already responsible for 3–4% of global electricity use, with projections showing further escalation driven by artificial intelligence (AI) and XR [ 9 , 8 ] . The unique demands of Metaverse environments amplify these challenges beyond traditional distributed computing scenarios. Unlike conventional applications, Metaverse platforms require real-time processing of multi-modal sensory data, continuous synchronization of shared virtual states across thousands of concurrent users, and ultra-low latency responses to maintain immersion [ 2 ] . These requirements translate to computational workloads that are not only massive in scale but also highly dynamic and unpredictable, making traditional static resource allocation strategies fundamentally inadequate. To address these intertwined challenges, we propose MetaFed , a decentralized, intelligent orchestration framework tailored for federated Metaverse computing. Our approach integrates three foundational innovations specifically designed for Metaverse infrastructural needs: • A multi-agent reinforcement learning (MARL) -based orchestration engine that dynamically adapts resource allocation based on system states, sustainability metrics, and privacy policies. MARL-based dynamic resource allocation is critically important for maintaining fluid user experiences under the constantly shifting user loads and interaction demands characteristic of Metaverse environments [ 3 ] . • A privacy-by-design federated learning (FL) architecture incorporating homomorphic encryption and differential privacy mechanisms to protect user data and model gradients. In Metaverse environments where sensitive biometric and behavioral data are continuously processed [ 7 ] , FL enhanced with homomorphic encryption and differential privacy serves as an indispensable privacy shield. • Carbon-aware scheduling strategies that leverage real-time energy grid data to prioritize low-emission computation, minimizing environmental footprint without compromising performance. Given the massive energy footprint potential of Metaverse infrastructure [ 9 ] , carbon-aware scheduling becomes a necessity rather than an option for ensuring environmental sustainability of this transformative technology. Thus, MetaFed aims to provide a robust response to the unique needs of Metaverse environments, enabling scalable, private, and sustainable orchestration across heterogeneous edge and cloud infrastructures. II Related Work II-A Federated Learning and Distributed AI FL has become the de facto framework for distributed machine learning (ML) without raw data centralization [ 10 ] . Techniques such as FedProx [ 11 ] , SCAFFOLD [ 12 ] , and FedNova [ 13 ] have attempted to address statistical heterogeneity and slow convergence. More recent works explore personalization [ 14 ] and byzantine robustness [ 15 ] . However, these approaches primarily focus on training efficiency and client reliability, without considering holistic orchestration or energy implications. II-B Resource Management in Edge Computing Edge computing facilitates latency-sensitive services by distributing compute close to data sources [ 16 , 17 ] . Resource orchestration strategies include dynamic task offloading [ 18 ] , bandwidth-aware scheduling [ 19 ] , and load balancing [ 20 ] . Reinforcement learning (RL)-based techniques [ 21 ] and game-theoretic models [ 22 ] have shown promise in adaptive orchestration. However, they often optimize single objectives, such as latency or throughput, and rarely incorporate privacy or sustainability constraints. II-C Sustainable and Green AI The energy cost of large-scale AI has led to the rise of Green AI, advocating for transparency in reporting energy use and promoting eco-efficient design [ 9 , 23 ] . Tools and methods for carbon-aware job scheduling [ 24 ] , renewable integration [ 25 ] , and geographical load balancing [ 26 ] have been proposed. Recent studies explore energy-efficient FL [ 27 ] and sustainable ML pipelines [ 28 ] , but lack integration with real-time orchestration or privacy protection. II-D Privacy-Enhancing Technologies in Federated Systems Maintaining user privacy in decentralized settings is critical. Techniques like differential privacy [ 29 ] , homomorphic encryption [ 30 ] , and secure aggregation protocols [ 31 ] offer varying trade-offs between computational overhead and privacy guarantees. While federated systems have begun integrating these methods, combining them with dynamic orchestration and environmental constraints remains underexplored. II-E Computing for Metaverse Infrastructures Recent surveys highlight the gaps in computing paradigms for the Metaverse [ 1 , 32 ] . While studies address immersive networking [ 33 ] , mobile edge integration [ 34 ] , and rendering acceleration [ 2 ] , few offer holistic solutions that jointly optimize performance, privacy, and sustainability. Existing Metaverse infrastructure proposals typically focus on single-objective optimization, either maximizing rendering performance, minimizing latency, or reducing costs, without considering the complex interplay between computational efficiency, user privacy, and environmental impact [ 35 ] . This fragmented approach fails to address the fundamental challenge that Metaverse platforms must simultaneously deliver exceptional user experiences while respecting privacy regulations and sustainability constraints. Therefore, MetaFed distinguishes itself by providing comprehensive, multi-objective optimization that treats performance, privacy, and sustainability as equally critical design constraints rather than optional considerations. Our proposed framework addresses this key research gap by offering the first principled approach to holistic Metaverse infrastructure orchestration. In summary, the current landscape in the literature offers valuable contributions in FL, privacy, and sustainable computing, but fails to comprehensively address the triadic challenge of intelligent, private, and green resource orchestration in federated Metaverse systems. MetaFed fills this gap with a principled, multi-objective orchestration architecture. III Methodology MetaFed operationalizes a three-pronged orchestration pipeline tailored to federated Metaverse environments. It