A newly proposed hierarchical adaptive federated learning framework improves model training accuracy and operational sustainability across large-scale Internet of Things networks. Researchers introduced the architecture to tackle persistent operational hurdles that arise when deploying distributed machine learning across decentralized smart devices. Traditional federated training protocols often fail to maintain balance across complex networks where participant devices experience differing physical operating constraints.
Standard distributed learning implementations such as FedAvg and FedProx focus almost exclusively on resolving statistical heterogeneity across local datasets. However, these conventional frameworks largely ignore crucial physical constraints, including residual device energy, varying wireless link quality, and bandwidth fluctuations. In massive Internet of Things deployments, disregarding these system-level limitations frequently leads to unstable convergence patterns and prematurely depletes the overall operating lifetime of the participating edge nodes.
To overcome these performance bottlenecks, the research team developed a hierarchical adaptive federated learning architecture combined with a self-attention VGG-16 network and a resource-aware aggregator. Designated as HAFL-SAVGG16, the system introduces a specialized weighted FedProx aggregation mechanism that departs from uniform update approaches. The aggregator dynamically recalibrates client participation weights at every communication round according to residual battery power, channel signal-to-noise ratio, and local data sample volumes.
Integrating self-attention mechanisms into the core VGG-16 architecture substantially enhances global contextual modeling capabilities across decentralized edge devices. This architectural enhancement allows edge devices to capture critical inter-device feature dependencies more effectively. Consequently, the network preserves learning robustness even when operating within noisy wireless environments and managing highly heterogeneous data distributions across diverse hardware components.
Extensive simulation benchmarks demonstrated that the HAFL-SAVGG16 framework achieved model accuracy rates of up to 98 percent within 20 communication rounds. Throughout the training lifecycle, the system reliably kept overall training loss below the 0.02 threshold. Simulation results also confirmed marked reductions in communication latency and operational energy consumption when compared against established baselines, including FedAvg, standard FedProx, SCAFFOLD, and FedNova.
The adaptive weighting mechanism maintained system stability under challenging experimental conditions featuring dynamic client availability and signal-to-noise ratio degradations spanning 10 to 40 percent. Under these volatile channel conditions, the proposed weighted aggregation mitigated convergence slowdowns by 3 to 12 percent and prevented accuracy drops of 5 to 9 percent. These empirical findings demonstrate the framework’s scalability and robustness for deploying sustainable, high-capacity intelligence across modern connected environments.
Subject of Research: Resource-aware hierarchical adaptive federated learning for Internet of Things networks
Article Title: A Novel Weighted FedProx Aggregator for Optimizing Resource Management in IoT Networks with HAFL-SAVGG16 and Global Aggregation
Article References: Prasad, S. K., & Ko, Y.-B. (2027). A Novel Weighted FedProx Aggregator for Optimizing Resource Management in IoT Networks with HAFL-SAVGG16 and Global Aggregation. Journal of Network and Systems Management, 35(1), Article 5. https://doi.org/10.1007/s10922-026-10095-w
Image Credits: AI Generated
DOI: 10.1007/s10922-026-10095-w
Keywords: federated learning, Internet of Things, resource management, FedProx, VGG-16, self-attention, HAFL-SAVGG16, energy efficiency, signal-to-noise ratio, client weighting, system heterogeneity, wireless networks, edge computing, training convergence, FedAvg, distributed optimization
News Source: Veronica Carney. (November 21, 2026). Adaptive Federated Learning Framework Boosts IoT Network Efficiency and Training Stability. Scienmag.



