Efficient UAV Swarm-Based Multitask Federated Learning With Dynamic Task Knowledge Sharing

doi:10.1109/JIOT.2026.3672872

Efficient UAV Swarm-Based Multitask Federated Learning With Dynamic Task Knowledge Sharing

2026-03-12

0

PRE

AI

DOI：10.1109/JIOT.2026.3672872

Original

Original request for help

Save

Abstract

En 中文

Uncrewed aerial vehicle (UAV) swarms are extensively used in emergency communications, area monitoring, and disaster relief. Their operations are coordinated by control centers, making them well-suited for federated learning (FL) frameworks. However, current UAV FL methods ignore the rich information contained in UAV images and the potential of using a single dataset to accomplish multiple tasks. For instance, in disaster relief scenarios, images acquired by UAVs can support tasks like crowd detection, road passability analysis, and disaster impact assessment. These tasks exhibit time-varying demands and may have potential correlations. To meet these requirements, this article introduces two core mechanisms: a dynamic task attention mechanism to evaluate task importance for efficient resource allocation, and a task affinity (TA) metric to capture intertask correlations for knowledge sharing. Building on these innovations, we propose FedDya, a novel UAV swarm-based one-dataset multitask FL framework, where ground emergency vehicles (EVs) collaborate with UAVs to accomplish multiple tasks leveraging a single dataset. To optimize resource allocation, we formulate a two-layer optimization problem to jointly optimize UAV transmission power, computation frequency, bandwidth allocation, and UAV-EV associations. For the inner problem, we derive closed-form solutions for transmission power, computation frequency, and bandwidth allocation and apply the block coordinate descent method for optimization. For the outer problem, a novel two-stage algorithm is designed to determine optimal UAV-EV associations. Furthermore, theoretical analysis reveals a tradeoff between UAV energy consumption violation and multitask performance, characterized by an $\mathcal {O}((V)^{1/2},1/V)$ relationship. Extensive simulation results further validate the effectiveness of the proposed scheme.

Keywords:

Knowledge sharing

one-dataset multitask federated learning (FL)

resource allocation and uncrewed aerial vehicle (UAV)-emergency vehicle (EV) association

task attention mechanism

Efficient UAV Swarm-Based Multitask Federated Learning With Dynamic Task Knowledge Sharing

Abstract

Journal

Organization