Decentralized Rank Scheduling for Energy-Constrained Multi-Task Federated Fine-Tuning in Edge-Assisted IoV Networks

2026-04-16 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Internet of Things (IoT) & Connected Devices, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

Researchers from Sun Yat-sen University propose a hierarchical federated fine-tuning framework for Internet of Vehicles (IoV) systems, designed to adapt foundation models to diverse tasks in energy-constrained edge environments. This framework coordinates roadside units (RSUs) and vehicles to enable resource-aware and mobility-resilient learning. It utilizes Low-Rank Adaptation (LoRA) and introduces a decentralized, energy-aware rank adaptation mechanism formulated as a constrained multi-armed bandit problem. A novel UCB-DUAL algorithm is developed for adaptive exploration under per-task energy budgets, achieving provable sublinear regret. Evaluated on a large-scale IoV simulator using real-world trajectories, the approach demonstrates superior accuracy-efficiency trade-off, reducing latency by over 24% and improving average accuracy by more than 2.5% compared to baselines like HomoLoRA and HetLoRA, while also significantly lowering CUDA memory usage.

Key takeaway

For research scientists developing AI solutions for dynamic edge environments like IoV, this framework offers a robust strategy for multi-task federated fine-tuning. You should consider adopting a hierarchical optimization approach with adaptive LoRA rank selection, particularly the UCB-DUAL algorithm, to manage heterogeneous resources and client mobility effectively. This can lead to significant improvements in accuracy, latency, and energy efficiency, crucial for real-time intelligent services in smart cities.

Key insights

A hierarchical federated fine-tuning framework optimizes LoRA rank selection for IoV, balancing accuracy, latency, and energy.

Principles

• Adaptive rank selection balances performance and cost.
• Hierarchical optimization improves resource distribution.
• Decentralized learning can achieve sublinear regret.

Method

The framework uses a two-level optimization: inter-task (server-side) for energy budget allocation based on task difficulty and utilization, and intra-task (client-side) for LoRA rank selection via a UCB-DUAL algorithm.

In practice

• Implement LoRA for efficient model adaptation.
• Use UCB-DUAL for decentralized resource allocation.
• Simulate with real-world trajectories for robust evaluation.

Topics

• Decentralized Rank Scheduling
• Federated Fine-Tuning
• Internet of Vehicles
• Low-Rank Adaptation
• UCB-DUAL Algorithm

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• Edge-AI-Driven Learning-to-Rank for Decentralized Task Allocation in Circular Smart Manufacturing — Aligning learning objectives with decentralized decision structures is critical for effective task allocation in smart manufacturing.
• PreLort: Prefix-Nested LoRA for Federated Fine-Tuning under Rank Heterogeneity — PreLort enables robust federated LoRA fine-tuning for LLMs by hierarchically structuring adapter ranks to manage client heterogeneity.
• Efficient Federated RLHF via Zeroth-Order Policy Optimization — Par-S^2ZPO offers efficient federated RLHF for edge devices using zeroth-order optimization and binary perturbation.
• Scheduling in a changing world: Maximizing throughput with time-varying capacity — New algorithms optimize non-preemptive job scheduling in dynamic cloud environments with fluctuating machine capacity.
• Hierarchical Advantage Weighting for Online RL Fine-Tuning of VLAs from Sparse Episode Outcomes — HABC improves VLA fine-tuning by hierarchically balancing viability and efficiency objectives with intervention-aware credit assignment.
• Towards Pareto-Optimal Tool-Integrated Agents with Pareto Ranking Policy Optimization — ParetoPO optimizes tool-integrated LLMs for superior accuracy-efficiency trade-offs using a two-stage multi-objective framework.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.