Recently, Federated Learning (FL) has revolutionized the processing and analysis of vast volumes of data generated by wireless devices, effectively overcoming the traditional cloud computing constraints within Internet of Things (IoT) networks. However, practical challenges arise as data on edge devices dynamically changes, necessitating continuous learning capabilities known as Federated Continual Learning (FCL). One key challenge in FCL is the issue of catastrophic forgetting, which refers to preserving the training performance on old data while training on new data. While common strategies involve retaining a subset of old data to mitigate the issue, privacy concerns limit this approach, and the balance between emphasis on new and old data during the training process remains inadequately studied. To address the above challenges, we propose an Adaptive Federated Continual Learning (AdapFCL) method in heterogeneous environment, which eliminates the need for episodic memory in federated settings. Specifically, the server employs a Deep Convolutional Generative Adversarial Network (DCGAN) model with a data-free knowledge distillation technique, which enables the server to learn representations of old data and generate synthetic data involving only global model. Then clients perform local training by utilizing new data and synthetic data instead of storing old data. Furthermore, we quantify the degree of forgetting on old data for each client, allowing for adaptive adjustment of emphasis weights for old and new data during the training process. Simulation results validate that the proposed method can achieve superior average test accuracy while maintaining communication efficiency compared with baselines, especially in highly heterogeneous data scenarios.

Autonomous underwater vehicles (AUVs) play a crucial role in data collection for underwater acoustic sensor networks (UWASNs). The limited capacity of individual AUV and the need for low-latency data collection necessitate the deployment of AUV swarms to achieve efficient and secure cooperative data collection. However, most existing works assume prior knowledge of sensor node locations, which is impractical in real-world AUV networks. Additionally, continuous data collection needs to be considered due to the sustained operation of sensors and cluster head replacement. To address these challenges, we propose a target uncertainty map assisted data collection scheme for AUV swarms based on the multiagent proximal policy optimization (MAPPO) algorithm. Specifically, the target uncertainty map is established by leveraging current and past search and collection results, guiding the AUV swarm to prioritize areas with higher probabilities of containing sensor nodes. Moreover, a digital pheromone mechanism incorporating repulsive and attractive pheromones is designed to establish an artificial potential field for adjusting the target uncertainty map. To further enable a comprehensive exploration of unknown environments, we introduce the Age of Information (AoI) as an indicator. Additionally, we consider the energy consumption associated with data collection to strike a balance between collection and energy efficiency, and derive a lower bound on the policy improvement achieved by the MAPPO algorithm. Simulation results have validated that the proposed scheme has a superior performance compared to the baselines, achieving an approximately 15% increase in the collection rate while reducing the energy consumption of data collection and AoI as well.

The development of Unmanned Aerial Vehicles (UAVs) offers new prospects for emerging applications in the Industrial Internet of Things (IIoT) networks. With the assistance of Digital Twin (DT), a real-time understanding of physical entities can be constructed for dynamic perception and decision-making. However, DT modeling requires distributed data aggregation, resulting in privacy disclosure and communication burden. Therefore, we propose the digital twin edge network by integrating the DT technology and edge computing, which leverages an over-the-air computation enabled federated learning architecture for an efficient and secure DT model construction. Specifically, we propose a heterogeneity-aware and energy-conscious device scheduling mechanism, considering the update importance, channel condition, and computation capacity based on a probabilistic scheduling framework. To enhance energy efficiency, we introduce a virtual queue to track the difference between the cumulative energy consumption and budget. Additionally, we design a low-complexity scheduling algorithm to solve the optimization problem. Simulation results validate the superiority of our proposed mechanism in improving the test accuracy and energy efficiency in a heterogeneous and energy-constrained environment. Moreover, the proposed mechanism demonstrates significant advantages when employed to highly heterogeneous datasets, and exhibits a certain level of robustness to mapping errors arising from the utilization of DT technique.

Recent years have envisioned the widespread adoption of machine learning (ML) in unmanned aerial vehicle (UAV) swarms for task execution. However, it is hard for the traditional centralized ML approaches to be applied in UAV swarms due to the large latency, communication cost, and privacy disclosure when transmitting the raw data. As an alternative, over-the-air computation (AirComp)-enabled federated learning (FL) is expected as a communication-efficient solution by harnessing the interference. However, the benefit of AirComp is at the cost of compromised training performance due to the channel distortion caused by the fading channel and noise and straggler issues resulting from the aligned parameters. In addition, the limited energy budget and dynamic environment incorporating the mobility characteristic of UAVs and time-varying channel conditions make these issues more complex. To solve problems aforementioned, this work proposes an energy-and-communication-efficient device scheduling scheme for AirComp-enabled FL system in the UAV swarms. Specifically, we first derive the optimality gap to characterize the impact of channel distortion and device selection on training performance. Then based on this result, we formulate an optimization problem to minimize the optimality gap by scheduling an appropriate number of competent following UAVs in each round, considering the transmission and computation energy consumption as well. Simulation results validate that the proposed scheme can achieve superior training performance in terms of test accuracy compared with baselines, and shows robustness with the increasing scale of UAV swarm.

In recent years, autonomous underwater vehicles (AUVs) have been widely applied to collect data in underwater acoustic sensor networks (UWASNs). Limited by the capacity of a single AUV, as well as the low-latency requirement of data collection, the intelligent swarm consisting of multiple AUVs is expected to execute the secure and efficient data collection tasks in a cooperative manner. However, most of the existing works assumed that the locations of sensor nodes are already known, which is impractical in a real AUV network. In addition, the security issues are not well considered in underwater searching and transmission tasks. To improve the searching efficiency in an unknown underwater area where locations of sensor nodes cannot be obtained precisely, this work proposes a data collection scheme via a target uncertainty map based multi-agent reinforcement learning algorithm for AUV swarms. Specifically, the target uncertainty map is established based on the current and past searching and collection results, which can guide the AUV swarm to search the areas with higher probabilities to find sensor nodes waiting for data collection. Moreover, to mitigate the potential security risk of data leakage, we design a multi-agent deep deterministic policy gradient (MADDPG) algorithm for each AUV in the swarm to make its searching and data collection strategies through a manner of centralized training with distributed execution. Simulation results validate that the proposed scheme can achieve a high collection rate with low energy consumption. In addition, the security referring to data protection can be also guaranteed in AUV swarms.

To satisfy the expected plethora of computation-heavy applications, federated edge learning (FEEL) is a new paradigm featuring distributed learning to carry the capacities of low-latency and privacy-preserving. To further improve the efficiency of wireless data aggregation and model learning, over-the-air computation (AirComp) is emerging as a promising solution by using the superposition characteristics of wireless channels. However, the fading and noise of wireless channels can cause aggregate distortions in AirComp enabled federated learning. In addition, the quality of collected data and energy consumption of edge devices may also impact the accuracy and efficiency of model aggregation as well as convergence. To solve these problems, this work proposes a dynamic device scheduling mechanism, which can select qualified edge devices to transmit their local models with a proper power control policy so as to participate the model training at the server in federated learning via AirComp. In this mechanism, the data importance is measured by the gradient of local model parameter, channel condition and energy consumption of the device jointly. In particular, to fully use distributed datasets and accelerate the convergence rate of federated learning, the local updates of unselected devices are also retained and accumulated for future potential transmission, instead of being discarded directly. Furthermore, the Lyapunov drift-plus-penalty optimization problem is formulated for searching the optimal device selection strategy. Simulation results validate that the proposed scheduling mechanism can achieve higher test accuracy and faster convergence rate, and is robust against different channel conditions.

Artificial intelligence (AI) is expected as a revo-lutionary technology to be widely used in Internet-of- Things (IoT) networks for computationally intensive tasks. However, the traditional centralized training framework imposes large latency, network burdens and high risk of privacy disclosure. As a promising distributed solution, federated learning involves the collaborative model training among edge devices, with the orchestration of a server to carry the capacities of low-latency and privacy preservation for AI -driven networks. To further improve the communication efficiency, over-the-air computation (AirComp) is capable of computing while transmitting data by exploiting the superposition property of wireless channels to harness the interference. However, gradient aggregation suffers from channel distortion induced by channel fading and noise, which may degrade the training performance. Moreover, it is beneficial to schedule the informative edge devices in federated learning under limited energy resources. In this work, we propose a dynamic device scheduling scheme for AirComp enabled federated learning systems. In this scheme, a proper number of qualified edge devices with channel inversion based power control are scheduled to participate the model training, where local updates diversity, channel condition and energy consumption are exploited jointly. Inspired by the Lyapunov drift-plus-penalty method, we formulate the optimization problem to attain the device selection strategy. Simulation results validate that the proposed scheme can achieve a close-to-optimal test accuracy with fast convergence rate, and present good performance of robustness under different channel conditions.