MOON-DPAP: Model-Contrastive Federated Learning with Differential Privacy and Adaptive Pruning

Jiaming Su

2025

Abstract

Recent years have seen a surge in research on distributed data processing and privacy protection due to the quick growth of big data and artificial intelligence technology. Federated learning, as a distributed collaboration framework with privacy protection, has attracted much attention due to its application potential. However, in practical applications, it faces challenges such as heterogeneous data distribution, high communication overhead, and insufficient privacy protection, and algorithm improvements are urgently needed to improve performance and adaptability. This study proposed an improved federated learning algorithm Model Contrastive Federated Learning-Differential Privacy and Adaptive Pruning (MOON-DPAP), which improved the efficiency, accuracy, and privacy protection capabilities of federated learning by introducing dynamic pruning technology, dropout, dynamic adjustment of differential privacy, and hyperparameter optimization. Experiments show that MOON-DPAP outperforms FedAvg, SCAFFOLD, MOON, and FedDyn in multiple performance indicators. In heterogeneous data scenarios, it shows higher accuracy and stability. In scalability tests, the algorithm performance remains superior even when the number of clients increases. Privacy protection tests verify its security and practicality. MOON-DPAP provides an innovative solution to the challenges of federated learning in performance improvement and privacy protection, laying the foundation for its practical application.

Paper Citation

in Harvard Style

Su J. (2025). MOON-DPAP: Model-Contrastive Federated Learning with Differential Privacy and Adaptive Pruning. In Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE; ISBN 978-989-758-765-8, SciTePress, pages 82-88. DOI: 10.5220/0013679100004670

in Bibtex Style

@conference{icdse25,
author={Jiaming Su},
title={MOON-DPAP: Model-Contrastive Federated Learning with Differential Privacy and Adaptive Pruning},
booktitle={Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE},
year={2025},
pages={82-88},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013679100004670},
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 - MOON-DPAP: Model-Contrastive Federated Learning with Differential Privacy and Adaptive Pruning
SN - 978-989-758-765-8
AU - Su J.
PY - 2025
SP - 82
EP - 88
DO - 10.5220/0013679100004670
PB - SciTePress