MadFed: Enhancing Federated Learning with Marginal-data Model Fusion.

As the demand for intelligent applications at network edge grows, so does need effective federated learning (FL) techniques. However, FL often relies on non-identically and non-independently distributed local datasets across end devices, which could result in considerable performance degradation. Prior solutions, such as model-driven approaches based knowledge distillation, meta-learning, transfer learning, have provided some reprieve. their suffers under heterogeneous highly skewed data distributions. To address these challenges, this study introduces MArginal Data fusion FEDerated Learning (MadFed) approach, a groundbreaking of model- data-driven methodologies. By utilizing marginal data, MadFed mitigates distribution skewness, improves maximum achievable accuracy, reduces communication costs. Furthermore, demonstrates that can significantly improve even with minimal entries, single entry. For instance, it provides up to 15.4% accuracy increase 70.4% cost savings when combined established Conversely, relying solely methodologies poor performance, especially datasets. Significantly, extends its effectiveness various algorithms offers unique method augment label sets thereby enhancing utility applicability real-world scenarios. The proposed approach is not only efficient but also adaptable versatile, promising broader application potential widespread adoption field.