Federated learning (FL) enables deployment of smart applications, where the intelligent model is trained with distributed data to achieve high accuracy and generalization capabilities. However, resource-constrained terminal devices are unable to train a global model at scale, leading to device dropouts and reduced data utilization. To achieve effective FL among heterogeneous models, we propose the edge learning based trusted heterogeneous federated learning scheme (ELTFL), which allows clients with different heterogeneous models to participate in FL through the application of knowledge distillation (KD). Specifically, ELTFL utilizes integrated distillation to build a heterogeneous multi-model training architecture for FL, which downloads global models to edge servers while preserving heterogeneous models of terminal devices. The edge server utilizes a small set of datasets along with probabilistic outputs from terminal devices to train the global model, and performs global model aggregation in the cloud. In addition, to safeguard against gradient leakage attacks, we introduce a lightweight iterative masking technique between edge servers and terminal devices to ensure the privacy and reliability of intermediate information. Extensive experiments on the CIFAR10 and CIFAR100 datasets demonstrate that ELTFL performs well in terms of accuracy, robustness to data heterogeneity, communication overhead, and privacy protection.

In the realm of edge computing, the analog aggregation based Federated Learning Over the Air (FLOA) emerges as a promising technology, offering heightened communication efficiency and privacy provisions. This approach involves concurrent transmission and aggregation, where edge devices (workers) collectively upload their local updates to a Parameter Server (PS) through shared time-frequency resources. The PS then obtains averaged updates only but not the individual local ones, reducing latency and communication costs. However, this simultaneous process exposes FLOA to vulnerabilities, particularly Byzantine attacks. Addressing this concern, we introduce an innovative framework utilizing Unmanned Aerial Vehicles (UAVs) to assist in a heterogeneous FLOA. This framework mitigates the impact of Byzantine attacks while preserving the advantages of over-the-air computation for efficient federated learning. The UAVs engage in over-the-air computation, collecting gradients from local workers and aggregating them. Subsequently, the UAVs transmit these aggregated gradients to the PS in different time slots. Robust aggregation techniques are applied at the PS to combine updates from UAVs. To enhance the robustness of over-the-air transmissions from workers to UAVs, we propose a power control policy. The resilience of the proposed framework to attacks is demonstrated through the derived expected convergence rate, validated by experiments on real datasets.

Unmanned Aerial Vehicles (UAVs) are promising for their agile flight capabilities, allowing them to carry out tasks in various complex scenarios. The efficiency and accuracy of UAV operations significantly depend on high-precision positioning technology. However, the existing positioning techniques often struggle to achieve accurate position estimates in conditions of Non-Line-Of-Sight (NLOS). To address this challenge, we propose a novel high-precision UAV positioning method based on MultiLayer Perceptron (MLP) integrating Ultra-WideBand (UWB) and Inertial Measurement Unit (IMU) technologies, which can acquire centimeter-level high-precision location estimation. In the method, we simultaneously extract key features from channel impulse responses and state space of UAV for training an MLP model, which can not only reduce error of UWB signals from dynamically flying UAV to anchor in NLOS environments, but also adapt to the diverse environment settings. Specifically, we respectively apply the anchor node assisted position calibration method and cooperative positioning techniques to the dynamic flying UAVs for solving the issues of UWB signal being blocked and lost. We conduct extensive real-world experiments to demonstrate the effectiveness of our approach. The results show that the median positioning errors of UAV in hovering and flight are 6.3 cm and within 20 cm, respectively.

Edge Computing (EC) pushes computational capability to the Terrestrial Devices (TDs), providing more efficient and faster computing solutions. Unmanned Aerial Vehicles (UAVs) equipped with EC servers can be flexibly deployed, even in complex terrains, to provide mobile computing services at all times. Meanwhile, UAVs can establish an air-to-ground line-of-sight link with TDs to improve the quality of communication link. However, the UAV-to-TD link may be obstructed by ground obstacles such as buildings or trees, leading to sub-optimal data transmission rates. To surmount this issue, Reconfigurable Intelligent Surfaces (RISs) emerge as a promising technology capable of intelligently reflecting signals to enhance communication quality between UAVs and TDs. In this paper, we consider the RISs-assisted multi-UAVs collaborative edge Computing Network (RUCN) in urban environment, in which we study the computational offloading problem. Our goal is to maximize the overall energy efficiency of UAVs by jointly optimizing the flight duration and trajectories of UAVs, and the phase shifts of RISs. It is worth noting that this problem has been formally established as NP-hard. Therefore, we propose the Deep Deterministic Policy Gradients based UAV Trajectory and RIS Phase shift optimization algorithm (UTRP-DDPG) to solve this complex optimization challenge. The results of extensive numerical experiments show that the proposed algorithm outperforms the other benchmark algorithms under various parameter settings. Specially, the UTRP-DDPG algorithm improves the UAV energy efficiency by at least 2% compared to DQN algorithm.

This paper studies two scheduling games on identical batching-machines with activation cost, where each game comprises $n$ jobs being processed on $m$ identical batching-machines. Each job, as an agent, chooses a machine (or, more accurately, a batch on a machine) for processing in order to minimize its disutility, which is comprised of its machine’s load and the activation cost it shares. Based on previous results, we present the Mixed strategy Nash Equilibria (MNE) for some special cases of the two games. For each game, we first analyze the conditions for the nonexistence of Nash equilibrium, then provide the MNE for the conditions, and offer the efficiency of MNE (mixed price of anarchy).