The large-scale model (LSM) can handle large-scale data and complex problems, effectively improving the intelligence level of urban intersections. However, the traffic conditions at intersections are becoming increasingly complex, so the intelligent intersection LSMs (I2LSMs) also need to be continuously learned and updated. The traditional cloud-based training method incurs a significant amount of computational and storage overhead, and there is a risk of data leakage. The combination of edge artificial intelligence and federated learning provides an efficient and highly privacy protected computing mode. Therefore, we propose a hierarchical hybrid distributed training mechanism for I2LSM. Firstly, relying on the intelligent intersection system for cloud-network-terminal integration, we constructed an I2LSM hierarchical hybrid distributed training architecture. Then, we propose a hierarchical hybrid federated learning (H2Fed) algorithm that combines the advantages of centralized federated learning and decentralized federated learning. Further, we propose an adaptive compressed sensing algorithm to reduce the communication overhead. Finally, we analyze the convergence of the H2Fed algorithm. Experimental results show that the H2Fed algorithm reduces the communication overhead by 21.6% while ensuring the accuracy of the model.


In the past decades, with the widespread implementation of wireless networks, such as the Internet of Things, an enormous demand for designing relative algorithms for various realistic scenarios has arisen. However, with the widening of scales and deepening of network layers, it has become increasingly challenging to design such algorithms when the issues of message dissemination at high levels and the contention management at the physical layer are considered. Accordingly, the abstract medium access control (absMAC) layer, which was proposed in 2009, is designed to solve this problem. Specifically, the absMAC layer consists of two basic operations for network agents: the acknowledgement operation to broadcast messages to all neighbors and the progress operation to receive messages from neighbors. The absMAC layer divides the wireless algorithm design into two independent and manageable components, i.e., to implement the absMAC layer over a physical network and to solve higher-level problems based on the acknowledgement and progress operations provided by the absMAC layer, which makes the algorithm design easier and simpler. In this study, we consider the implementation of the absMAC layer under jamming. An efficient algorithm is proposed to implement the absMAC layer, attached with rigorous theoretical analyses and extensive simulation results. Based on the implemented absMAC layer, many high-level algorithms in non-jamming cases can be executed in a jamming network.