SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (32) · 27574-27582
DOI: 10.1609/aaai.v40i32.39977 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Mingkun Yang und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Split Federated Learning is a system-efficient federated learning paradigm that leverages the rich computing resources at a central server to train model partitions. Data heterogeneity across silos, however, presents a major challenge undermining the convergence speed and accuracy of the global model. This paper introduces Step-wise Momentum Fusion (SMoFi), an effective and lightweight framework that counteracts gradient divergence arising from data heterogeneity by synchronizing the momentum buffers across server-side optimizers. To control gradient divergence over the training process, we design a staleness-aware alignment mechanism that imposes constraints on gradient updates of the server-side submodel at each optimization step. Extensive validations on multiple real-world datasets show that SMoFi consistently improves global model accuracy (up to 7.1%) and convergence speed (up to 10.25x). Furthermore, SMoFi has a greater impact with more clients involved and deeper learning models, making it particularly suitable for model training in resource-constrained contexts.
Zitieren
APA
Yang, M., Zhu, R., Wang, Q., & Yang, J. (2026). SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data. Proceedings of the AAAI Conference on Artificial Intelligence, 40(32), 27574-27582. https://doi.org/10.1609/aaai.v40i32.39977
Vancouver
Yang M, Zhu R, Wang Q, Yang J. SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(32):27574-27582. doi:10.1609/aaai.v40i32.39977
BibTeX
@article{Yang2026,
author = {Yang, Mingkun and Zhu, Ran and Wang, Qing and Yang, Jie},
title = {SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {32},
pages = {27574--27582},
doi = {10.1609/aaai.v40i32.39977}
}
EUVIMED