MetaFed: Advancing Privacy, Performance, and Sustainability in Federated Metaverse Systems

TL;DR

MetaFed introduces a decentralized federation framework for Metaverse computing that jointly optimizes performance, privacy, and sustainability. By combining a multi-agent reinforcement learning orchestrator with privacy-preserving federated optimization (Enhanced FedAvg, FedProx, and homomorphic encryption) and carbon-aware scheduling, it achieves higher accuracy while substantially reducing CO2 emissions on MNIST and CIFAR-10. The results demonstrate substantive improvements in environmental impact with minimal increases in communication rounds, suggesting MetaFed as a scalable approach for privacy-conscious and energy-aware Metaverse infrastructures. This work offers a practical blueprint for eco-friendly, private, and high-performance distributed Metaverse systems with broad implications for edge-cloud orchestration, green AI, and federated learning adoption.

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.

Figures (5)

• Figure 1: Accuracy–emission trade-off for MNIST: Comparison of federated learning models based on classification accuracy and per-round CO2 emissions.
• Figure 2: Accuracy over communication rounds on the MNIST dataset for MetaFed and baseline federated learning models.
• Figure 3: Accuracy over communication rounds on the CIFAR-10 dataset for MetaFed and baseline federated learning models.
• Figure 4: Accuracy–emission trade-off for CIFAR-10: Comparison of federated learning models based on classification accuracy and per-round CO2 emissions.
• Figure 5: MetaFed architecture showing interaction between resource providers, federated agents, and carbon-aware scheduling.