Improving Communication Efficiency in Federated Learning via Metaheuristic-Based Learning Algorithms

Federated learning is a distributed machine learning approach that utilizes multiple devices' coordination for training purposes. Nevertheless, distributed computing tends to result in slow training. It is important to properly tune the hyperparameters, which include the learning rate, the number of local training rounds, and the size of the local data volume to optimize the training speed. In this paper, the FedIAVOA algorithm is proposed, which introduces an improved IAVOA algorithm to select the optimal combination of hyperparameters to improve the federated system's communication efficiency. Experimental results on MNIST and CIFAR-10 datasets show that the FedIAVOA algorithm outperforms the FedAvg algorithm in terms of model accuracy and communication efficiency, achieving a balance between model accuracy and communication cost.