Multi-agent dual federated learning for intelligent task offloading in hierarchical edge–fog–cloud 6G networks

The rapid emergence of sixth-generation (6G) wireless networks is driving the need for intelligent task offloading frameworks capable of simultaneously satisfying stringent latency, energy efficiency, scalability, and privacy requirements. Existing task offloading approaches are constrained by centralized learning architectures, binary offloading decisions, inadequate handling of heterogeneous non-independent and identically distributed (non-IID) data, and limited awareness of wireless interference dynamics, resulting in suboptimal resource utilization in hierarchical Edge–Fog–Cloud (EFC) environments. To address these challenges, this study proposes MADFL-6G, a Multi-Agent Dual Federated Learning framework for intelligent task offloading in hierarchical 6G networks — ‘Dual’ referring to MWFA’s two-component aggregation weighting, which jointly combines data-volume proportionality and KL-divergence-based distribution-similarity penalization (Sect. 3.5, Eq. 9). The proposed framework integrates distributed Dueling Deep Q-Network (Dueling DQN) agents with Prioritized Experience Replay (PER), a Mobility-Weighted Federated Averaging (MWFA) strategy for robust aggregation under heterogeneous data distributions, an interference-aware seven-dimensional state representation, and a five-action adaptive partial task offloading mechanism that dynamically distributes computational workloads across edge, fog, and cloud resources. The task offloading process is formulated as a set of coupled, independently learning agents operating within a shared, non-stationary Edge–Fog–Cloud environment, incorporating stochastic wireless channel conditions and queueing dynamics to accurately model realistic 6G environments. Extensive experiments conducted on more than 200,000 real task execution records demonstrate that MADFL-6G achieves an average latency of 61 ms, energy consumption of 0.31 J, and a 98.7% task success rate, outperforming conventional federated deep Q-learning by 28.2%, 31.1%, and 2.6%, respectively, while substantially exceeding heuristic and centralized baselines. Additional analyses, including scalability evaluation, component ablation, statistical significance testing, robustness under non-IID settings, and SHAP-based explainability, confirm the effectiveness, interpretability, and reliability of the proposed framework. These findings demonstrate that MADFL-6G provides a scalable, privacy-aware, raw-data-preserving, and explainable solution for intelligent resource management in emerging 6G applications, including autonomous transportation, industrial Internet of Things, smart cities, and extended reality services.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-16
DOI
https://doi.org/10.1038/s41598-026-78296-6
Primary Topic
IoT and Edge/Fog Computing
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multi-agent dual federated learning for intelligent task offloading in hierarchical edge–fog–cloud 6G networks

S. Muthu Vijaya Pandian, P. Sundaravadivel, E. S. Vinothkumar, B. Ambika
4 links
Scientific Reports
IoT and Edge/Fog Computing
article

Multi-agent dual federated learning for intelligent task offloading in hierarchical edge–fog–cloud 6G networks

S. Muthu Vijaya Pandian, P. Sundaravadivel, E. S. Vinothkumar, B. Ambika
article en

Abstract

The rapid emergence of sixth-generation (6G) wireless networks is driving the need for intelligent task offloading frameworks capable of simultaneously satisfying stringent latency, energy efficiency, scalability, and privacy requirements. Existing task offloading approaches are constrained by centralized learning architectures, binary offloading decisions, inadequate handling of heterogeneous non-independent and identically distributed (non-IID) data, and limited awareness of wireless interference dynamics, resulting in suboptimal resource utilization in hierarchical Edge–Fog–Cloud (EFC) environments. To address these challenges, this study proposes MADFL-6G, a Multi-Agent Dual Federated Learning framework for intelligent task offloading in hierarchical 6G networks — ‘Dual’ referring to MWFA’s two-component aggregation weighting, which jointly combines data-volume proportionality and KL-divergence-based distribution-similarity penalization (Sect. 3.5, Eq. 9). The proposed framework integrates distributed Dueling Deep Q-Network (Dueling DQN) agents with Prioritized Experience Replay (PER), a Mobility-Weighted Federated Averaging (MWFA) strategy for robust aggregation under heterogeneous data distributions, an interference-aware seven-dimensional state representation, and a five-action adaptive partial task offloading mechanism that dynamically distributes computational workloads across edge, fog, and cloud resources. The task offloading process is formulated as a set of coupled, independently learning agents operating within a shared, non-stationary Edge–Fog–Cloud environment, incorporating stochastic wireless channel conditions and queueing dynamics to accurately model realistic 6G environments. Extensive experiments conducted on more than 200,000 real task execution records demonstrate that MADFL-6G achieves an average latency of 61 ms, energy consumption of 0.31 J, and a 98.7% task success rate, outperforming conventional federated deep Q-learning by 28.2%, 31.1%, and 2.6%, respectively, while substantially exceeding heuristic and centralized baselines. Additional analyses, including scalability evaluation, component ablation, statistical significance testing, robustness under non-IID settings, and SHAP-based explainability, confirm the effectiveness, interpretability, and reliability of the proposed framework. These findings demonstrate that MADFL-6G provides a scalable, privacy-aware, raw-data-preserving, and explainable solution for intelligent resource management in emerging 6G applications, including autonomous transportation, industrial Internet of Things, smart cities, and extended reality services.

Scientific Reports
Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology (IN), Anna University, Chennai (IN), Saveetha University (IN)
Industry, innovation and infrastructure
IoT and Edge/Fog Computing
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon