Reducing computational cost in federated ensemble learning via rank-one matrix

As distributed environments have developed and data privacy has become more important, federated learning attracts more attentions. Federated learning is the method to train deep learning models without data exchange in distributed environments. However, in a Non-Independent and Identically Distributed data environment, performance degradation occurs in the federated learning environment. We want solve this problem by federated ensemble learning but computational cost problem occurs in edge devices. To solve this problem, we proposes FedRE. FedRE reduces computational cost on ensemble model via rank-one matrix and using shared weights to minimize performance degradation when Non-IID situations. As a result of the experiments, SVHN and CIFAR-10 image classification tasks showed high accuracy compared to the previous federated learning methods.