@article{MC21295CE,
title = "A Study on Distributed Learning Algorithm for Heterogeneous Client Settings in Computing Capabilities",
journal = "The Journal of Korean Institute of Communications and Information Sciences",
year = "2025",
issn = "1226-4717",
doi = "10.7840/kics.2025.50.2.195",
author = "Ji-Hyun Ryu, Heecheol Yang",
keywords = "Distributed learning, Federated learning, Split learning",
abstract = "In this paper, we address a distributed learning system with a server and clients with heterogeneous computational capabilities. We propose a new distributed learning algorithm that combines split learning with ordered dropout, enabling clients with limited and heterogeneous computing capabilities to participate in training. This approach allows all clients, even with different client-model sizes, to contribute to the improvement of the global model’s performance. We conduct experiments on image classification using ResNet50 on the CIFAR-10 dataset, examining classification performance given the number of clients and the distribution of the dataset. With heterogeneous client settings in computational capacities, simulation results demonstrate that all clients with various client-side model sizes effectively contribute to global model training."
}