Research on Distributed Machine Learning Training Based on Serverless Computing Platforms

Xinyue Duan, Taimin Rong

2025

Abstract

Currently, since 2023, with the further development of artificial intelligence, the requirements for the efficiency and accuracy of federated learning have also increased. In distributed computing environments, the heterogeneity of hardware configurations and data distributions across different devices poses challenges to model training. Designing a federated learning algorithm that can adapt to heterogeneous environments and effectively perform knowledge transfer is a key research challenge. To address the construction of multi-task distributed models with minimal communication resource consumption, this paper presents the advanced Knowledge Transfer-Personalized Federated Learning (KT-pFL) algorithm. KT-pFL is based on the training and improvement of the Heterogenous Federated Learning via Model Distillation (FedMD) algorithm, and the core logic of the algorithm is briefly described. This paper proposes improvement methods for the KT-pFL algorithm based on the reproduced algorithm, including the addition of normalization layers and dynamic adaptive adjustment of the learning rate. These improvements can increase the accuracy of the algorithm by approximately 2% when training on the CIFAR-10 dataset. Finally, a brief analysis and outlook on potential future research directions for knowledge transfer are provided.

Paper Citation

in Harvard Style

Duan X. and Rong T. (2025). Research on Distributed Machine Learning Training Based on Serverless Computing Platforms. In Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE; ISBN 978-989-758-765-8, SciTePress, pages 111-116. DOI: 10.5220/0013679500004670

in Bibtex Style

@conference{icdse25,
author={Xinyue Duan and Taimin Rong},
title={Research on Distributed Machine Learning Training Based on Serverless Computing Platforms},
booktitle={Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE},
year={2025},
pages={111-116},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013679500004670},
isbn={978-989-758-765-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE
TI - Research on Distributed Machine Learning Training Based on Serverless Computing Platforms
SN - 978-989-758-765-8
AU - Duan X.
AU - Rong T.
PY - 2025
SP - 111
EP - 116
DO - 10.5220/0013679500004670
PB - SciTePress