Decentralized Rank Scheduling for Energy-Constrained Multi-Task Federated Fine-Tuning in Edge-Assisted IoV Networks
Summary: arXiv:2508.09532v2 Announce Type: replace-cross
Abstract
Federated fine-tuning has emerged as a promising approach for adapting foundation models (FMs) to diverse downstream tasks in edge environments. In Internet of Vehicles (IoV) systems, enabling efficient and low-latency multi-task adaptation is particularly challenging due to client mobility, heterogeneous resources, and intermittent connectivity.
This paper proposes a hierarchical federated fine-tuning framework that coordinates roadside units (RSUs) and vehicles to support resource-aware and mobility-resilient learning across dynamic IoV scenarios. Leveraging Low-Rank Adaptation (LoRA), we introduce a decentralized, energy-aware rank adaptation mechanism formulated as a constrained multi-armed bandit problem.
Key Contributions
- Development of a hierarchical federated fine-tuning framework for IoV systems.
- Implementation of a decentralized, energy-aware rank adaptation mechanism.
- Formulation of the adaptation problem as a constrained multi-armed bandit problem.
- Introduction of the UCB-DUAL algorithm for adaptive exploration within energy budgets.
Methodology
A novel UCB-DUAL algorithm is developed to enable adaptive exploration under per-task energy budgets, achieving provable sublinear regret. This methodology is particularly significant as it allows for flexible adjustments based on the energy constraints of various tasks.
To evaluate our method, we constructed a large-scale IoV simulator based on real-world trajectories, capturing dynamic participation, RSU handoffs, and communication variability. This simulator provided a robust environment to assess the performance of our proposed framework.
Results
Extensive experiments demonstrate that our approach achieves the best accuracy-efficiency trade-off among all baselines. Key findings include:
- Reduction in latency by over 24% compared to existing methods.
- Improvement in average accuracy by more than 2.5%.
- Enhanced adaptability to dynamic IoV environments, ensuring consistent performance amidst client mobility and resource variability.
Conclusion
In conclusion, our proposed hierarchical federated fine-tuning framework significantly advances the capabilities of IoV systems in multi-task learning scenarios. By addressing the challenges of energy constraints and mobility, this work lays the groundwork for future research in decentralized learning frameworks. The UCB-DUAL algorithm, in particular, offers a promising approach for adaptive exploration in energy-constrained environments, marking a step forward in optimizing federated learning in edge-assisted IoV networks.