CAIROS: Adaptive Control of Federated Learning in Wireless Networks

• Lucas Airam C. de Souza UFRJ / INRIA Saclay
• Nadjib Achir INRIA Saclay
• Miguel Elias M. Campista UFRJ
• Luís Henrique M. K. Costa UFRJ

DOI: https://doi.org/10.5753/sbrc.2026.19690

Abstract

Vehicular Federated Learning (VFL) is applied to the training of AI models to ensure user data privacy. However, clients exhibit greater variation in the communication channel than in static scenarios due to high client mobility, exceeding the round response timeout. This reduces system performance, as updates from straggler clients are discarded if they exceed the transmission timeout for the round. This work proposes CAIROS, a strategy for model training in vehicular learning that allows each client to estimate its network and computing conditions through an LSTM model. Based on this estimate, the client decides whether to continue training or to send the calculated parameters early to avoid a timeout. The results show that CAIROS, compared to FedAvg, reduces the incidence of discarded updates due to timer expiration in VFL by up to 38%, increasing the accuracy of the trained models by up to 25%.

References

Beutel, D. J. et al. (2020). Flower: A Friendly Federated Learning Research Framework. arXiv preprint arXiv:2007.14390.

Chatzoulis, D. et al. (2023). 5G V2X Performance Comparison for Different Channel Coding Schemes and Propagation Models. Sensors, 23(5):2436.

Clancy, J. et al. (2024). Wireless Access for V2X Communications: Research, Challenges and Opportunities. Communications Surveys & Tutorials.

De Souza, L. A. C. et al. (2024). AutoMHS-GPT: Automated Model and Hyperparameter Selection with Generative Pre-Trained Model. Em CloudNet, páginas 1–8. IEEE.

de Souza, L. A. C. et al. (2025). TOFL: Time Optimized Federated Learning. Em SBSeg. SBC.

Gong, J., Liu, W., Pei, M., Wu, C. e Guo, L. (2022). ResNet10: A Lightweight Residual Network for Remote Sensing Image Classification. Em ICMTMA, páginas 975–978.

Horváth, S. et al. (2021). FjORD: Fair and Accurate Federated Learning under heterogeneous targets with Ordered Dropout. Em Advances in Neural Information Processing Systems, volume 34, páginas 12876–12889.

Kim, G. et al. (2025). FedAWT: Adaptive Federated Learning via Dynamic Epoch Adjustment for Heterogeneous Clients in Ad-Hoc Networks. Internet of Things, 34:101771.

Krizhevsky, A. et al. (2009). Learning Multiple Layers of Features from Tiny Images.

Liu, J. et al. (2021). Adaptive Asynchronous Federated Learning in Resource-Constrained Edge Computing. Transactions on Mobile Computing, 22(2):674–690.

Mahmoud, S. et al. (2025). Speed Up Federated Learning in Heterogeneous Environments: A Dynamic Tiering Approach. Internet of Things Journal, 12(5).

McMahan, B. et al. (2017). Communication-efficient Learning of Deep Networks from Decentralized Data. Artificial Intelligence and Statistics, páginas 1273–1282.

Nguyen, J. et al. (2022). Federated Learning with Buffered Asynchronous Aggregation. Em International Conference on Artificial Intelligence and Statistics, páginas 3581–3607. PMLR.

Ono, S. e Nakao, A. (2025). Adaptive Timing Control of Parameter Aggregation in Vehicular Federated Learning. Em ICCE, páginas 1–6. IEEE. ISSN: 2158-4001.

Park, J. et al. (2022). AMBLE: Adjusting Mini-Batch and Local Epoch for Federated Learning with Heterogeneous Devices. JPDC, 170:13–23.

Patanè, R. et al. (2024). Can Vehicular Cloud Replace Edge Computing? Em WCNC. IEEE.

Thomaz, G. A. et al. (2025). AGATA – Arquitetura para Gerenciamento Automático de Tarefas de Aprendizado Federado. Em SBRC, páginas 121–128. SBC.

Wang, Z. et al. (2022). Asynchronous Federated Learning over Wireless Communication Networks. Transactions on Wireless Communications, 21(9):6961–6978.

Wu, W. et al. (2020). SAFA: A Semi-Asynchronous Protocol for Fast Federated Learning with Low Overhead. Transactions on Computers, 70(5):655–668.

Xu, C., Qu, Y., Xiang, Y. e Gao, L. (2023). Asynchronous Federated Learning on Heterogeneous Devices: A Survey. Computer Science Review, 50:100595.

Xu, Z. et al. (2021). Helios: Heterogeneity-aware Federated Learning with Dynamically Balanced Collaboration. Em DAC, páginas 997–1002. IEEE.

Yang, E. et al. (2018). An Adaptive Batch-Orchestration Algorithm for the Heterogeneous GPU Cluster Environment in Distributed Deep Learning System. Em International Conference on Big Data and Smart Computing (BigComp), páginas 725–728. IEEE.

Zhu, Q. et al. (2021). 3GPP TR 38.901 Channel Model. Em The Wiley 5G Ref: The Essential 5G Reference Online, páginas 1–35. Wiley Press Hoboken, NJ, USA.