Improved Knowledge Mapping in Heterogeneous Network Using Enhanced Federated Learning

Edge networks consist of devices that analyze different data types and models. A knowledge map can be developed in device scheduling using homogeneous or heterogeneous models. In this study, a Federated learning algorithm is used to minimize communication overhead by distributing device information in batches, ensuring that all devices have an equal opportunity. The federated learning method is utilized in this study to allocate spectrum between primary and secondary users by categorizing them using standard parameters such as bandwidth utilization, energy, and a novel metric known as signal-to-noise ratio. Here, the spectrum is allocated using the deep learning technique. However, in other algorithms, global loss minimization was not considered for the model analysis. Fortunately, in this study, model analysis was carried out using deep learning architectures such as Convolution Neural Networks for feature extraction, pooling layers for downsampling, and accessing the performance using evaluation metrics. The findings indicated that knowledge mapping could also be improved by improving the model.