FedAux：Accelerating Traditional Federated Learning Process
Accelerating Traditional Federated Learning Process and Improving Global Model Accuracy by Training Shared Data as Auxiliary Models with Transfer Learning and Shallow/Deep Layer Theories of Deep Learning.
Model Architecture Diagram：
Description： Currently, hybrid federated learning algorithms have not been effectively utilizing shared data nor mitigating the challenges of data and device heterogeneity in federated learning. Therefore, a more effective solution for sharing data and coordinating local data is proposed. FedAux leverages transfer learning and shallow/deep layer theories of deep learning to train shared data as auxiliary models and transfer their general knowledge to device models, accelerating the traditional federated learning process and improving the accuracy of the global model.This predictor is developed using tensorflow and keras under the Python 3.6+ environment. It involves a specified number of devices for federated learning, and the algorithm updates the global model and device local models iteratively. It should be noted that during the implementation of the code, the predictor refers to third-party open-source resources such as the Hybrid-FL protocol and feature transfer theory of deep neural networks.