Journal article · 2026 · Open access
Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning
IEEE Transactions on Mobile Computing, 25(11), 20900–20916 · Published 20 July 2026 · DOI 10.1109/tmc.2026.3715159
In brief
“Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning” is a 2026 journal article by Bowen Li and 7 co-authors, published in IEEE Transactions on Mobile Computing. OpenAlex counts 0 citations as of 19 November 2026. It is open access under a CC BY licence: the full text is free to read at IEEE Transactions on Mobile Computing.
- Citations
- 0
- References
- 56
- Authors
- 8
Abstract
Federated fine-tuning combines the capabilities of pre-trained models with personalized data on mobile devices. However, deploying pre-trained models imposes high memory and communication burdens on resource-limited devices. Although Low-Rank Adaptation (LoRA) reduces these burdens, existing approaches overlook the inherent heterogeneity of mobile devices.Spatial heterogeneity(uneven device capabilities) causes delays when fast devices wait for slow counterparts.Temporal heterogeneity(on-device resource dynamics) includes network fluctuations, resulting in bandwidth drops that lead to system-wide bottlenecks. Our insight reveals that LoRA rank (inner dimension of LoRA matrices) acts as a controllable variable to balance model accuracy and time efficiency. Spatial and temporal heterogeneity can be mitigated by adaptively scheduling LoRA ranks based on device and network conditions. Based on this insight, we propose FedHLoRA:FederatedHeterogeneousLoRAFine-Tuning. By establishing the correlation between time and rank, FedHLoRA tackles dual heterogeneity through real-time rank allocation to minimize synchronization delays. Furthermore, it employs layer-wise aggregation to minimize server-side latency. Experiments confirm that FedHLoRA reduces time costs by 58% (ResNet) and 9% (DistilBERT) versus baselines while reaching identical accuracy. To the best of our knowledge,FedHLoRA is the first method to dynamically adapt LoRA ranks based on heterogeneous user resources in federated fine-tuning.
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Li, B., Jiang, J., Lu, R., Lin, Y., Zhang, W., Yang, W., Wang, F., & Wang, Z. (2026). Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning. IEEE Transactions on Mobile Computing, 25(11), 20900–20916. https://doi.org/10.1109/tmc.2026.3715159
MLA 9
Li, Bowen, et al. “Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning.” IEEE Transactions on Mobile Computing, vol. 25, no. 11, 2026, pp. 20900–20916. https://doi.org/10.1109/tmc.2026.3715159.
Chicago
Li, Bowen, Jingyan Jiang, Rongwei Lu, Yixi Lin, Wei Zhang, Wei Yang, Fangxin Wang, and Zi Wang. 2026. “Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning.” IEEE Transactions on Mobile Computing 25 (11): 20900–20916. https://doi.org/10.1109/tmc.2026.3715159.
Harvard
Li, B. et al. (2026) ‘Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning’, IEEE Transactions on Mobile Computing, 25(11), pp. 20900–20916. doi:10.1109/tmc.2026.3715159.
IEEE
B. Li et al., “Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning,” IEEE Transactions on Mobile Computing, vol. 25, no. 11, pp. 20900–20916, 2026, doi: 10.1109/tmc.2026.3715159.
BibTeX
@article{li2026adaptive,
title = {{Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning}},
author = {Li, Bowen and Jiang, Jingyan and Lu, Rongwei and Lin, Yixi and Zhang, Wei and Yang, Wei and Wang, Fangxin and Wang, Zi},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {11},
pages = {20900--20916},
doi = {10.1109/tmc.2026.3715159},
url = {https://doi.org/10.1109/tmc.2026.3715159}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Topic and keywords
Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- federated learning
- LoRa
- low-rank adaptation
- heterogeneous devices
- spatial heterogeneity
- temporal heterogeneity
- synchronization delay
- mobile devices
Work with this paper
Most-cited papers in Computer Science
- ImageNet classification with deep convolutional neural networksAlex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton · Communications of the ACM · 2017 · 114,726 citations
- Long Short-Term MemorySepp Hochreiter, Jürgen Schmidhuber · Neural Computation · 1997 · 109,097 citations
- A Mathematical Theory of CommunicationClaude E. Shannon · Bell System Technical Journal · 1948 · 91,736 citations
- Deep learningYann LeCun, Yoshua Bengio, Geoffrey E. Hinton · Nature · 2015 · 90,226 citations
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksShaoqing Ren, Kaiming He, Ross Girshick and 1 more · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2016 · 74,143 citations
Questions about this paper
How do I cite this paper in APA style?
Li, B., Jiang, J., Lu, R., Lin, Y., Zhang, W., Yang, W., Wang, F., & Wang, Z. (2026). Adaptive Heterogeneous Model Scheduling for Federated LoRA Fine-Tuning. IEEE Transactions on Mobile Computing, 25(11), 20900–20916. https://doi.org/10.1109/tmc.2026.3715159
How many times has this paper been cited?
0 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.
Is this paper free to read?
Yes. It is open access at IEEE Transactions on Mobile Computing: sign in and open the full text from this page.
Who wrote this paper?
Bowen Li, Jingyan Jiang, Rongwei Lu, Yixi Lin, Wei Zhang, Wei Yang, Fangxin Wang, Zi Wang.
What is the DOI of this paper?
10.1109/tmc.2026.3715159, permanently at https://doi.org/10.1109/tmc.2026.3715159. Sign in and open it from this page.
Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.