A differential privacy-enhanced federated edge learning framework, designated as DECFel, achieves higher model accuracy and lower loss than existing approaches such as DP-FedAvg and FedProx in experimental benchmark evaluations. The framework was developed to deliver an improved trade-off between privacy protection and model performance during collaborative training across distributed edge devices. In distributed edge environments, federated learning is implemented to train models collaboratively while maintaining data locality on local hardware and reducing communication overhead across networks.
Deploying federated learning across edge environments introduces several critical operating challenges that the new framework directly confronts. Transmitting model updates between edge devices and central coordinators risks privacy leakage, leaving systems vulnerable to curious or malicious servers. Concurrently, strict resource constraints on edge hardware can lead to substantial performance degradation during training cycles. To mitigate these competing concerns, the researchers developed a joint optimization formulation designed specifically around the operational delay and energy constraints inherent to edge devices.
The formulated joint optimization problem targets the simultaneous minimization of global model loss alongside a weighted composite measure comprising privacy leakage, execution delay, and device energy consumption. Because federated edge learning must adaptively safeguard local data without exhausting hardware resources, DECFel incorporates an adaptive mechanism that injects calibrated noise into local model updates. This optimization problem couples multiple competing objectives, requiring a specialized calculation strategy to determine suitable training parameters without imposing prohibitive overhead on resource-limited devices.
To resolve this coupled optimization problem efficiently, DECFel implements a two-phase framework that balances computational efficiency with model performance. In the first phase, a particle swarm optimization algorithm utilizing lightweight approximate evaluation rapidly explores the search space to identify optimal hyperparameters. In the second phase, precise federated training proceeds using analytically derived optimal noise values.
Subject of Research: Differential privacy and optimization in federated edge learning
Article Title: Adaptive differentially private federated edge learning under delay and energy constraints
Article References: Zhang, G., Cui, C., Man, Z., & Hu, W. (2027). Adaptive differentially private federated edge learning under delay and energy constraints. Cluster Computing, 30(1), Article 30. https://doi.org/10.1007/s10586-026-06672-y
Image Credits: AI Generated
DOI: 10.1007/s10586-026-06672-y
Keywords: federated learning, edge computing, differential privacy, DECFel, particle swarm optimization, delay constraints, energy consumption, privacy leakage, distributed devices, data locality, communication overhead, hyperparameter optimization, DP-FedAvg, FedProx, adaptive noise injection, model updates, global loss minimization
News Source: Veronica Carney. (November 19, 2026). Adaptive Differential Privacy Framework Improves Federated Learning Performance on Edge Devices. Scienmag.



