Monday 10 March 2025
In recent years, the rapid growth of Artificial Intelligence (AI) and Machine Learning (ML) has enabled the development of numerous innovative applications in various fields. Among these advancements is Hierarchical Federated Learning over vehicle-edge-cloud architectures, which has gained significant attention due to its potential to revolutionize the way data is processed and analyzed.
In a recent study published in the IEEE Journal of Selected Areas in Communications, researchers from Xiamen University and other institutions presented a novel approach to tackle the challenges associated with Hierarchical Federated Learning. The proposed methodology, called Hybrid Evolutionary And gReedy allocaTion (HEART), aims to achieve timely model training while ensuring balanced training across various tasks.
To understand the significance of this research, it’s essential to grasp the concept of Hierarchical Federated Learning. In traditional federated learning settings, multiple devices or edge nodes learn from their local data and aggregate their models to create a global model. However, this approach can be inefficient and time-consuming when dealing with large-scale datasets or complex tasks.
In contrast, Hierarchical Federated Learning involves multiple levels of aggregation, where local devices learn from their own data and then send the aggregated models to edge nodes for further processing. Edge nodes, in turn, aggregate the received models and send them to cloud servers for global model training. This hierarchical architecture enables more efficient and scalable learning.
The researchers behind HEART recognized that one major challenge in Hierarchical Federated Learning is the need to balance task scheduling and model training priorities among edge nodes and vehicles. To address this issue, they proposed a hybrid approach combining Particle Swarm Optimization (PSO) and Genetic Algorithms (GA).
In their methodology, HEART first uses PSO to achieve balanced task scheduling by optimizing the allocation of tasks to edge nodes. This is followed by GA, which determines the training priority of assigned tasks on vehicles. By integrating these two optimization techniques, HEART ensures that tasks are scheduled efficiently and model training is optimized.
The researchers tested their methodology using real-world datasets and compared it with existing methods. The results showed significant improvements in terms of time efficiency, with HEART reducing overall non-task-training time by up to 51.7% compared to baseline methods.
Furthermore, the study demonstrated that HEART can effectively handle various scenarios, including different numbers of vehicles and data samples per vehicle. This adaptability makes it a promising solution for real-world applications.
Cite this article: “Hybrid Optimization Approach for Hierarchical Federated Learning in Vehicle-Edge-Cloud Architectures”, The Science Archive, 2025.
Artificial Intelligence, Machine Learning, Hierarchical Federated Learning, Vehicle-Edge-Cloud Architectures, Particle Swarm Optimization, Genetic Algorithms, Task Scheduling, Model Training, Edge Nodes, Cloud Servers
