Federated Learning (FL) confronts challenges such as resource constraints and unbalanced data distribution in the Space-Air-Ground Integrated Network (SAGIN). This paper proposes a Hierarchical Split Federated Learning (HSFL) framework considering satellite handoff and derives its upper bound of loss function affected by model splitting and data distribution. To minimize the weighted sum of training loss and latency, we formulate a joint optimization problem that integrates device association, model split layer selection, and resource allocation. We decompose the original problem into several subproblems, where an iterative optimization algorithm incorporating closed-form solutions and brute-force split point search is proposed. Simulation results demonstrate that the proposed algorithm can balance training efficiency and model accuracy for FL in SAGIN.

Space-air-ground integrated vehicular networks (SAGVN) play a crucial role in the 6G system, offering global coverage and ultra-wide-area broadband access. Meanwhile, advancements in artificial intelligence (AI) have led to a significant increase in model inference demands, which come with stringent latency requirements. In this paper, considering the intelligent services in SAGVN, we explore the co-inference problem and improve the inference efficiency by performing model splitting, vehicle association, and resource allocation. This problem is solved iteratively by decomposing it into multiple sub-problems. In particular, the model splitting problem is addressed by systematically searching for the optimal split points. Besides, the joint vehicle association and resource allocation problem are reformulated into a monotone submodular function without satellites, and then a low-complexity submodular optimization algorithm is proposed. To further utilize the computing power of the satellite, we further introduce a resource reallocation algorithm based on the existing optimization results. These two subproblems can iterate alternately until the optimization goal converges. Simulation results show that the proposed algorithm can achieve better latency performance in several inference tasks.

Space-Air-ground integrated vehicular network (SAGVN) can provide substantial advantages for the Internet of Vehicles (IoV) with broad coverage and long-distance communications. However, efficient task offloading in SAGVN is difficult due to the dynamic and multi-dimensional characteristics of IoV. In this paper, we address a task offloading problem in SAGVN, where unmanned aerial vehicles (UAVs) and low Earth orbit (LEO) satellites collaborate to offer mobile edge computing (MEC) services to vehicles. Our goal is to jointly design service placement and task offloading strategies for each UAV to minimize overall system latency, subject to mobility, coverage, energy, and bandwidth constraints. The problem can be reformulated as a multi-agent Markov decision process (MAMDP), where each vehicle and UAV can act as an agent, and the actions taken by agents correspond to the optimal UAV trajectory, subchannel selection, task partition ratio, and offloading destination. Then, we decompose the problem into three sub-problems of service placement, task offloading, and subchannel selection. Given the extensive observation and action space, an iterative best response multi-agent proximal policy optimization (IBR-MAPPO) algorithm is proposed. Finally, simulation results show that our approach converges rapidly and achieves lower execution delay than baseline algorithms.

Millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) is one of the most promising technologies from 5G-based Internet of Things (IoT) to future wireless communication-based IoT, which usually relies on beamforming codebooks for data transmission. However, traditional codebooks often consist of numerous narrow beams, which causes substantial training overhead. Although centralized machine learning-based methods can address this issue to some extent, they overlook minority IoT devices scattered across various areas, which is vital for the coverage equity of the environmental adaptive codebook and the optimal average achievable rate. To circumvent the problem, we propose a distributed learning (DL) framework for codebook design in mmWave massive MIMO systems with uneven user distribution. Specifically, the user channel set is first divided into subsets by pre-classification based on the power responses of the featured combining vectors from different subregions. Then, a novel DL architecture processes these subsets, each assigned to different baseband processing boards (BPBs) in building baseband units, alleviating the centralized machine learning burden on the active antenna unit (AAU) or its directly connected BPB. Meanwhile, the current algorithms lack the hardware perspective or only implement the inference stage of the model. Thus, we deploy an FPGA-adapted DL-based codebook training prototype that runs on FPGA, which fully explores the “Backward-While-Forward" strategy for data reuse in the forward and backward passes. Simulation validates the effectiveness of distributed learning. Notably, the FPGA implementation on the embedded-level board outperforms consumer-grade CPU and GPU in terms of both latency and energy efficiency.

The development of Industrial Internet of Things (IIoT) applications presents a critical challenge in terms of latency limitation, particularly considering the limited availability of resources that prevent a single fog device from fully executing large-scale computing tasks. In such scenarios, enabling distributed computing across multiple fog servers or collaborating with cloud servers holds promising potential. To improve the efficiency of task offloading while accounting for the crucial role of movable fog devices (e.g., robots and unmanned cars), we formulate a joint optimization problem as a partially observable Markov decision process (POMDP), incorporating offloading decisions, computing resource allocation, and trajectory optimization under constraints related to available resources and collision avoidance. Due to the nondeterministic polynomial-time hardness (NP-hardness) in the problems of task offloading and resource allocation, we reformulate a matroid-constrained submodular maximization problem and propose an iterative low-complexity algorithm to find solutions. Subsequently, extracting better solutions from submodular optimization, we propose a multiagent reinforcement learning (MARL)-based algorithm to solve the trajectory optimization problem for the movable fog devices acting as agents, making decisions based on their local observations. Finally, simulation results have validated that the proposed scheme has a superior performance compared to the baselines.

Federated learning (FL) framework enables user devices to collaboratively train a global model based on their local data sets without privacy leak. However, the training performance of FL is degraded when the data distributions of different devices are incongruent. Fueled by this issue, we consider a clustered FL (CFL) method where the devices are divided into several clusters according to their data distributions and are trained simultaneously. Convergence analysis is conducted, which shows that the clustered model performance depends on cosine similarity, device number per cluster, and device participation probability. Besides, to quantify the training performance, the utility of clustered model training is defined based on the analysis results. Then, aiming at optimizing the system utility, a joint problem of resource allocation and device clustering is formulated, which is solved by decoupling it into two subproblems. First, given the results of device clustering, a low-complexity iterative algorithm based on the convex optimization theory is proposed to make the bandwidth allocation and the transmit power control. Then, according to the individual stability, a coalition formation algorithm is proposed for the device clustering. Finally, the real-data experiments on the classification tasks (e.g., MNIST, CIFAR-10, and CIFAR-100) validate the results of convergence analysis and advantages of the proposed algorithm in terms of the test accuracy.

The Industrial Internet of Things (IIoT) is emerging as a promising technology that can accelerate the application of industrial intelligence to smart factories. Because of the sensitive nature of user data, federated learning (FL) which performs distributed machine learning while preserving data privacy, is leveraged to meet the accuracy and privacy requirements of IIoT end devices/clients. However, the unreliable communications in IIoT may result in possible single-point failures in the typical single-server FL framework, thereby negatively affecting the training efficiency. In this paper, we study on the client scheduling problem in a multi-server FL framework for the communication reliability and training efficiency improvement. Specifically, we focus on a semi-decentralized FL (SD-FL) framework, where edge servers and clients collaborate to train a shared global model through unreliable intra-cluster model aggregation and inter-cluster model consensus because of the model transmission error in client-server and server-server communication. Then, a client-server association optimization problem is formulated, with the objective of minimizing the global training loss. Resorting to the convergence analysis of SD-FL, the original problem is simplified and transformed into an integer nonlinear programming problem to guide us to design a high-efficiency client scheduling scheme. Finally, experimental results show that the proposed scheme significantly outperforms the baselines in terms of the test accuracy and training loss.

Integrated sensing and communication in 6G, particularly for air-ground surveillance using automatic dependent surveillance-broadcast (ADS-B) and multi-lateration (MLAT) systems, is gaining significant research interest. This letter investigates the problem of optimal anchor station selection for tracking aerial vehicles, and proposes a novel heuristic learning scheme termed as fishing net-like optimization (FNO). Specifically, we perform constrained random walk steps on a two-dimensional surface to optimize the initial anchor stations’ parameters. FNO also incorporates with new evaluation strategies and acceleration techniques to accelerate the convergence speed. Experimental results demonstrate that FNO can achieve better selection of the anchor stations, and the accuracy of the chosen MLAT can be improved by ten times or more with the anchors optimization.

Federated learning (FL) emerges as a privacy-preserving paradigm to effectively integrate edge computing for the implementation of deep learning-based vehicular applications. Nevertheless, the incentive mechanism for the vehicles to participate with varied learning tasks, has not been well explored yet. In this paper, for the training control among vehicles, a multi-task FL framework is investigated, where multiple edge servers collectively coordinate numerous vehicular models from different learning tasks. Aiming at motivating the vehicles to actively participate in training while improving the system utility of multiple learning tasks, a problem of multi-task contract design is formulated, and an iterative reward allocation algorithm is proposed based on the property of the convergence performance and the contract constraints. Extensive experiments validate that the proposed algorithm can achieve the balance between the system utility and the cluster utility, with higher test accuracy.

Federated learning (FL) framework enables user devices collaboratively train a global model based on their local datasets without privacy leak. However, the training performance of FL is degraded when the data distributions of different devices are incongruent. Fueled by this issue, we consider a clustered FL (CFL) method where the devices are divided into several clusters according to their data distributions and are trained simultaneously. Convergence analysis is conducted, which shows that the clustered model performance depends on cosine similarity, device number per cluster, and device participation probability. Then, aiming at optimizing the model training performance, a joint problem of resource allocation and device clustering is formulated, which is solved by decoupling it into two sub-problems. Specifically, a coalition formation algorithm is proposed for the device clustering sub-problem, and the sub-problem of bandwidth allocation and transmit power control is solved directly due to its convexity. Finally, simulation experiments are conducted on the MNIST dataset to validate the performance of the proposed algorithm in terms of test accuracy.

In this paper, we consider a hierarchical federated learning system and formulate a joint problem of edge aggregation interval control and time allocation to minimize the weighted sum of training loss and training latency. To quantify the learning performance, an upper bound of the average global gradient deviation, in terms of the edge aggregation interval, the time allocated for training, and the number of successfully participating devices, is derived. Then an alternative problem is formulated, which can be decoupled into two sub-problems and solved with two steps. In the first step, given the time allocation strategy, a relaxation and rounding method is proposed to optimize the edge aggregation interval. In the second step, with the results of the obtained edge aggregation interval and based on the convex optimization theory, an optimal time allocation can be evaluated. Simulation results show that the proposed scheme, compared to the benchmarks, can achieve higher learning performance with lower training latency.

Federated learning (FL) has become a promising solution to train a shared model without exchanging local training samples. However, in the traditional cloud-based FL framework, clients suffer from limited energy budget and generate excessive communication overhead on the backbone network. These drawbacks motivate us to propose an energy-aware hierarchical federated learning framework in which the edge servers assist the cloud server to migrate the local models from the clients. Then a joint local computing power control and client association problem is formulated in order to minimize the training loss and the training latency simultaneously under the long-term energy constraints. To solve the problem, we recast it based on the general Lyapunov optimization framework with the instantaneous energy budget. We then propose a heuristic algorithm, which takes the importance of local updates into account, to achieve a suboptimal solution in polynomial time. Numerical results demonstrate that the proposed algorithm can reduce the training latency compared to the scheme with greedy client association and myopic energy control, and improve the learning performance compared to the scheme in which the associated clients transmit their local models with the maximal power.