Adaptive UAV-Assisted Hierarchical Federated Learning: Optimizing Energy, Latency, and Resilience for Dynamic Smart IoT

TL;DR

This paper tackles the problem of energy- and delay-efficient UAV-assisted Hierarchical Federated Learning in dynamic IoT environments with intermittent connectivity. It introduces a joint optimization framework that co-optimizes learning configuration, bandwidth, device-to-UAV associations, UAV redeployment, and aggregator roles, formulating it as the NP-hard $\mathcal{P}_0$ and decomposing into $\mathcal{P}_1$, $\mathcal{P}_2$, and $\mathcal{P}_3$. Subproblems are solved via PALM-BLO for learning/communication, a TD3-based MD planning for adaptive device associations using KL-divergence based data-similarity, and a two-stage greedy method for UAV redeployment, with extensive experiments showing significant reductions in training time and energy while preserving or enhancing convergence under UAV dropouts. The approach, termed CEHFed, demonstrates practical impact for robust, energy-aware learning in distributed IoT settings and offers a path toward scalable, resilient autonomous aerial-edge learning systems.

Abstract

Hierarchical Federated Learning (HFL) extends conventional Federated Learning (FL) by introducing intermediate aggregation layers, enabling distributed learning in geographically dispersed environments, particularly relevant for smart IoT systems, such as remote monitoring and battlefield operations, where cellular connectivity is limited. In these scenarios, UAVs serve as mobile aggregators, dynamically connecting terrestrial IoT devices. This paper investigates an HFL architecture with energy-constrained, dynamically deployed UAVs prone to communication disruptions. We propose a novel approach to minimize global training costs by formulating a joint optimization problem that integrates learning configuration, bandwidth allocation, and device-to-UAV association, ensuring timely global aggregation before UAV disconnections and redeployments. The problem accounts for dynamic IoT devices and intermittent UAV connectivity and is NP-hard. To tackle this, we decompose it into three subproblems: \textit{(i)} optimizing learning configuration and bandwidth allocation via an augmented Lagrangian to reduce training costs; \textit{(ii)} introducing a device fitness score based on data heterogeneity (via Kullback-Leibler divergence), device-to-UAV proximity, and computational resources, using a TD3-based algorithm for adaptive device-to-UAV assignment; \textit{(iii)} developing a low-complexity two-stage greedy strategy for UAV redeployment and global aggregator selection, ensuring efficient aggregation despite UAV disconnections. Experiments on diverse real-world datasets validate the approach, demonstrating cost reduction and robust performance under communication disruptions.

Figures (11)

• Figure 1: A schematic of HFL architecture over smart IoT of our interest with energy-constrained UAVs and dynamic devices that move between various UAV's coverage areas.
• Figure 2: An overview of our decoupling framework for solving problem $\mathcal{P}_0$.
• Figure 3: A schematic of our TD3 framework.
• Figure 4: Performance comparisons in terms of test accuracy on MNIST and FaMNIST datasets using different models.
• Figure 5: Time cost of model training on MNIST and FaMNIST datasets upon having different models.

Theorems & Definitions (6)

• Remark 1: Departure and Arrival of UAVs
• Theorem 1
• Theorem 2
• Theorem 3
• Theorem 4
• Remark 2