In dynamic wireless environments, fluctuations in user distribution and channel conditions challenge the stability and efficiency of federated learning (FL) systems, especially in multi-UAV-assisted scenarios. UAV-assisted wireless FL has drawn great attention for its flexible deployment and distributed computing capabilities. However, centralized aggregation faces the risk of single-node failure at aggregation nodes, which can severely impact system reliability in dynamic environments. To address these issues, we propose a Decentralized Aggregation in UAV-assisted wireless Federated Learning (DAUFL) framework, which replaces centralized aggregation with a multi-node model aggregation mechanism to maintain stable operation and enhance system reliability under node failures. Furthermore, we develop an Adaptive Interference-aware Resource allocation and Deployment Optimization (AIRDO) algorithm. It first employs Constrained K-means for efficient UAV coverage, then applies Multi-Agent Reinforcement Learning (RL) for dynamic interference-aware resource allocation and UAV deployment optimization. Simulation results show that DAUFL achieves strong robustness and model performance under both Independent and Identically Distributed (IID) and Non-IID data distributions, maintaining over 90% and 80% accuracy even under a 50% node failure rate. The AIRDO algorithm further accelerates convergence, reduces training latency, and maintains stable performance with large-scale users.

The rapid development of the Internet of Medical Things (IoMT) has led to the deployment of sensors around the human body to collect physiological signals, enabling doctors to diagnose diseases and monitor users’ health information in real-time. However, collected physiological data often contains sensitive personal information that must be protected to ensure personal privacy. Data steganography provides an effective means of securing such data, and recent advancements in machine learning have further enhanced steganography capabilities for data security. In this paper, we propose a Dual-Adversarial Steganographic Generative Adversarial Network (DAS-GAN), an end-to-end reversible steganography framework for medical grayscale images. To ensure full usability of medical images after data extraction, a reversible steganography mechanism is integrated to recover the original image without any loss. Moreover, a dual-adversarial architecture is employed, leveraging adversarial properties from both spatial and transform domains to enhance data security and imperceptibility. The proposed DAS-GAN is deployed on a Raspberry Pi platform and evaluated using real-world medical datasets. Experimental results demonstrate that DAS-GAN effectively safeguards sensitive data while maintaining high visual fidelity and the diagnostic utility of medical images.

Integrated sensing and communication (ISAC) provides an effective solution for supporting both high-rate data transmission and environmental perception in connected and automated vehicles (CAVs). However, in urban environments, buildings frequently hinder signal propagation between the base station (BS) and vehicles, thereby compromising link reliability and weakening the effectiveness of radar-based sensing. To mitigate these challenges, intelligent reflecting surface (IRS) technology has been proposed to enhance CAV operations by improving the signal propagation environment. This study focuses on the ISAC system implemented in IRS-assisted CAV scenarios. In this system, the IRS facilitates the simultaneous transmission of communication and radar signals. By jointly optimizing the active beamforming matrix at the BS and the passive beamforming matrix at the IRS, the sum rate achievable by vehicular users is maximized, while ensuring compliance with the BS’s power budget, maintaining required radar detection quality, and adhering to the IRS phase configuration limits. In light of these considerations, an approach, namely Adaptive Optimization for Sensing and Forming (AOSF), is proposed, which employs the semidefinite relaxation (SDR) and successive convex approximation (SCA) algorithms to optimize the active beamforming matrix at the BS and employs the fractional programming algorithm to optimize the passive beamforming matrix at the IRS. The method decomposes the non-convex optimization problem into a series of manageable subproblems and employs an iterative solving approach to gradually refine the results and converge toward a satisfactory solution. Simulation results indicate that the proposed approach significantly improves achievable sum rate performance and reveals the trade-offs between communication and sensing.

The exponential proliferation of short video services has imposed severe strain on mobile backhaul links, necessitating intelligent proactive caching at the network edge to guarantee user Quality of Experience (QoE). Although proactive caching offers a promising avenue to alleviate congestion, its effectiveness is often constrained by the challenge of capturing rapid shifts in user interests and the intricate diffusion of content across social networks. To bridge this gap, we propose SC-STRL, a Social-Cognitive Spatiotemporal Reinforcement Learning framework that integrates temporal interest evolution with social structure analysis. Specifically, SC-STRL models user community influence to identify key opinion leaders and utilizes a Spatiotemporal Encoding Architecture involving a Temporal Interest Evolution Network (TIEN) and a Social Structure Embedding Module (SSEM) to capture dynamic demands. Building on these insights, we further devise a specialized Double Deep Q-Network (DDQN) agent augmented with Prioritized Experience Replay (PER), featuring a novel social-interactive reward mechanism explicitly designed to align caching decisions with content propagation dynamics. Extensive experiments on the KuaiRec dataset demonstrate SC-STRL's superiority, achieving performance improvements of 12.02% over the state-of-the-art benchmarks at a cache capacity of 400.

Low earth orbit satellite-assisted Internet of Vehicles (LEO SIoV) supports intelligent driving applications in areas without terrestrial network coverage. Virtual network embedding (VNE) demonstrates significant potential for optimizing resource utilization in LEO SIoV by orchestrating on-demand sharing of scarce resources while meeting tailored user requirements. However, conventional centralized VNE algorithms face significant challenges in load balancing efficiency and scalability constraints. To overcome these challenges, a load-aware semi-centralized VNE strategy designed for LEO SIoV is proposed, aiming to alleviate network load while maximizing network revenue. Within this strategy, a domain-partitioning-based load-aware semi-centralized VNE method is introduced to mitigate the load imbalance inherent in centralized VNE approaches. By dynamically selecting underutilized domains for each embedding operation, this method significantly improves load balancing and service request acceptance rates within LEO SIoV. Furthermore, to tackle scalability limitations, a deep reinforcement learning (D- RL) approach enhanced with an integrated Graph Attention Network and Sequence-to-Sequence module (iGATSeq) is proposed to handle dynamic topology changes and time-varying network requests. Extensive simulations demonstrate that our proposed method outperforms the compared baselines, achieving 24.17% faster convergence and 6.52% higher network revenue.

This paper proposes a secure space-air-ground in tegrated network (SAGIN) transmission scheme in which an unmanned aerial vehicle (UAV) equipped with reconfigurable holographic surfaces (RHS) antennas to tackle the challenges of ultra-long distances in ground-space (G2S) links and eaves dropping risks in wireless channels. Leveraging the holographic beamforming principle, the UAV can relay confidential G2S signals while transmitting interference as artificial noise toward eavesdroppers to counter interception. The closed-form expressions of the secrecy transmission rate for SAGIN are first derived. Then, an optimization strategy is designed under onboard energy consumption constraints, which jointly optimizes the beam radiation amplitude, 3D trajectory and power allocation, aiming to maximize the average secrecy rate (ASR) while satisfying the quality of service for legitimate users. A block coordinate descent algorithm is employed to decouple the original non convex problem into four subproblems, and feasible solutions are obtained through successive convex approximation. Simulation results validate the advantages of the proposed scheme, with the ASR significantly outperforming baseline schemes, such as those without holographic beamforming, horizon-only trajectory deployment, and without power control. The variations of the ASR with RHS element size and UAV transmit power are also analyzed. This research provides new theoretical insights and technical solutions for secure communication in SAGIN.

Low earth orbit (LEO) satellite networks, as integrated service systems, are typically divided into domains based on application functions, such as observation and communication. However, independent domain resources and the high dynamics of satellites make cross-domain resource interactions difficult to capture, posing challenges to large-scale satellite network resource coordination and degrading the quality of service (QoS) for task flows. To address this, a multi-level hypergraph (MLH) is introduced to represent domain resources across temporal, resource type and spatial dimensions. MLH consolidates similar resource features via hyperedges, reducing redundant connections. The cross-domain resource coordination problem is then modeled as a mixed-integer linear programming (MILP) problem to maximize the sum of the minimum priorities of scheduled tasks. Furthermore, by leveraging MLHs topological nesting, a multi-dimensional resource dual-level scheduling algorithm (MRDSA) is proposed, decomposing the problem into two subproblems solved using the hyperpath scheduling algorithm (HSA) and internal path scheduling algorithm (IPSA). Simulations demonstrate that the proposed method enhances QoS, reduces computational complexity and improves resource utilization ratio in LEO satellite networks.

Bitrate plays a crucial role in the transmission of semantic symbols, necessitating a trade-off between semantic information preservation and transmission efficiency. However, most existing Semantic Communication (SEMCOM) designs employ pre-fixed bitrate schemes that lack the flexibility to adapt to dynamic communication environments. To address this limitation, we propose a novel two-phase varibale bitrate control algorithm designed to dynamically adjust the bitrate, thereby achieving an optimal balance between semantic performance and communication resource consumption. In Phase I, we introduce a Deep Learning (DL) based Imitation Network to model the relationship between semantic performance and bitrate across varying channel conditions, enabling the calculation of environment-specific rewards. Subsequently, Phase II employs a deep reinforcement learning (DRL) agent to predict the optimal bitrate for each communication round through environmental exploration. Experimental results demonstrate that our approach yields at least 10% and 2.3% performance improvements over fixed-length bitrate systems under diverse user requirements and bit budgets.

Screen content images (SCIs), characterized by a mixture of discrete textual elements and continuous natural content, pose significant challenges for high-quality compression at low bitrates. To address this, we propose a novel joint framework that synergistically combines textual semantic representation and super-resolution (SR) techniques for efficient low-bitrate SCI compression. Our approach introduces a hybrid compression pipeline: textual content is extracted and encoded based on its semantic attributes, while the non-textual background undergoes pixel-wise removal and is compressed using a dedicated super-resolution strategy tailored for screen content. A key innovation is the pixel-wise textual content removal scheme, which leverages the Segment Anything Model (SAM) for precise text segmentation and the advanced LaMa inpainting model to seamlessly reconstruct the background, thereby preserving critical texture details. Furthermore, we design a specialized super-resolution network for the decompressed background, which effectively restores sharp edges and mitigates compression artifacts through global feature fusion and a tailored loss function. Extensive experiments on public SCID and SIQAD datasets demonstrate that our method achieves an average Bjontegaard-Delta (BD) rate reduction of 44.10% compared to the state-of-the-art semantic-based method TSA-SCC [1], with a maximum reduction reaching of 70.57%, significantly enhancing visual quality at low bitrates.

While multi-device cooperative task-oriented semantic communication (TOSC) enhances task performance through comprehensive information representation, it inevitably introduces redundancy, thereby increasing communication overhead. Existing redundancy elimination methods suffer from limitations in interpretability and coarse granularity, hindering the optimal utilization of communication resources. To this end, we propose a disentangled information bottleneck guided TOSC framework (DisenIB-TOSC). The framework first employs the basic IB for initial feature compression, then formulates a novel DisenIB specifically designed for multi-device cooperation inference, which enhances task performance and achieves interpretable feature disentanglement by separating features into common and private components, thereby establishing a theoretical foundation for redundancy identification. Subsequently, we derive differentiable, computationally tractable forms for both IB objectives by combining variational approximation, consistency constraints, and density ratio trick. Leveraging the disentangled features, we further design a feature importance-aware selective transmission strategy, DisenIB-TOSC-ST, which quantifies feature importance via mutual information estimation to dynamically discriminate and control redundant feature transmission. Experimental results on several tasks demonstrate that our method outperforms baselines in task performance while reducing communication costs, verifying the effectiveness and interpretability of feature disentanglement.

Computing Power Network (CPN) is a new paradigm that integrates communication, computing, and storage resources to provide services for tasks. However, tasks composed of non-independent subtasks have a preference for the resources required at each stage, which increases the difficulty of heterogeneous resource allocation and reduces the latency performance of CPN services. Motivated by this, this paper jointly optimizes the full-service cycle of tasks, including transmission, task partitioning, and offloading. First, the transmission bandwidth is dynamically configured based on delay sensitivity of tasks. Second, with the real-time information from edge resource clusters and state resource clusters in the network, the optimal partitioning for a computation task is derived. Third, personalized resource allocation schemes are customized for computation and storage tasks respectively. Finally, the impact of resource parameter configuration on the latency violation probability of CPN is revealed. Moreover, compared with the benchmark schemes, our proposed scheme reduces the network latency violation probability by up to 1.17 × in the same network setting.

Low earth orbit satellite-assisted Internet of Vehicles (LEO SloV) supports intelligent driving applications in areas beyond terrestrial network coverage. However, its long transmission distances and highly dynamic topology pose significant challenges to deterministic service assurance, such as degraded information freshness and low service request acceptance rates. To address these issues, this work proposes an information freshness-aware deterministic service assurance strategy for LEO SloV. The problem is formulated as a joint optimization of information freshness, service function chain (SFC) deployment cost, and information freshness fairness. A hierarchical deep reinforcement learning (HDRL) framework is developed, which integrates a graph attention network (GAT) for feature extraction and a gated recurrent unit (GRU)-based sequence-to-sequence (Seq2Seq) model as the policy network, enabling adaptive service request admission control and deterministic SFC deployment. Simulation results demonstrate that the proposed method outperforms baseline approaches, improving the service request acceptance rate by approximately 21.29%, reducing the average age of information by 15.96%, and lowering resource consumption by 4.75%.

Green computation has emerged as one of the key goals of cloud resource provisioning. The cloud resource clusters (CRCs) composed of heterogeneous performance servers are powered by uninterruptible power supplies (UPSs). However, due to the inherent nonlinear losses of UPSs, CRCs face challenges in computing resource provisioning with regard to sustainable energy consumption. Meanwhile, to alleviate the backlog in service queues, we propose a joint resource configuration and instance placement optimization problem that comprehensively models the energy consumption during instances processing in CRCs. Specifically, based on the length of the execution intervals of these two phases, this problem is decomposed into two subproblems in a dual-timescale framework. In a long timescale, we explore the temporal correlation of historical request data to pre-configure resources. Subsequently, the Lyapunov optimization method is employed to decompose the energy consumption problem of servers supported by each UPS into multiple subproblems across short timescale, while ensuring the stability of service queues. Furthermore, we use instance continuous relaxation to derive the optimal placement solution ideally, and design a double optimal gradient descent strategy for its practical implementation. Evaluation results demonstrate that the proposed strategy achieve measurable energy reduction in CRCs while maintaining flexibility during resource scaling.

As the healthcare industry continues to embrace digital transformation, the Internet of Medical Things (IoMT) emerges as a key technology. IoMT plays a critical role in revolutionizing healthcare delivery by enabling remote patient monitoring, personalized treatments, and efficient healthcare management. This survey offers a comprehensive overview of IoMT, elucidating its concepts and architectural framework. It explores its diverse applications and the challenges associated with its adoption. Additionally, it investigates key methods in lightweight, explainable artificial intelligence, discusses their applications in healthcare services, and outlines relevant evaluation metrics. Furthermore, it examines data security and privacy concerns in healthcare, presenting relevant methodologies. Finally, this review provides insights into the future of IoMT, considering existing challenges and opportunities for advancement.

Hierarchical federated learning (HFL) emerges as a promising solution for distributed systems in real-world IoT, compensating for the limitations of client and training data through multilevel aggregation. However, severe system heterogeneity, such as data heterogeneity, differentiated communication and computing power, can cause model drift and negatively impact system performance, posing challenges to existing HFL frameworks. To address this issue, we propose a mitigating edge heterogeneity method for HFL (MEH-Fed) in this paper. First, we design a new HFL framework, where the key idea is to introduce an initial update of the edge server side to reduce the impact of heterogeneity on system performance. Edge servers use synthetic datasets to guide the update direction of client models, facilitating model training and accelerating convergence. Then, we conduct a convergence analysis of the proposed method, establish a theoretical upper bound, and analyze the effects of key parameters. Finally, we experimentally evaluate the performance of the proposed MEH-Fed in terms of system performance and convergence rate. Experimental results validate that MEH-Fed achieves optimal performance in both accuracy and efficiency, establishing a new paradigm in HFL architecture design for heterogeneous IoT environments.

In Connected and Autonomous Vehicles (CAVs), traditional network architectures require vehicle users to access application services via Macro Base Stations (MBS) that are connected to the core network, leading to a significant delay. Edge caching leverages the storage resources of Small Base Stations (SBS) to pre-cache content, effectively reducing latency for vehicle content access. However, CAVs face constraints on spectrum and computing resources, and the Quality of Service (QoS) requirements vary across different application services. To address the varying tolerances for user latency and freshness, this study considers multiple content application services with diverse QoS requirements and measures data freshness using Age of Information (AoI). When a vehicle user initiates a service request, minimizing the associated cost while meeting latency and AoI constraints within limited resources has emerged as a key research problem. This paper introduces a Multi-Agent Reinforcement Learning (MARL) algorithm incorporating graph neural networks to optimize cache distribution and resource allocation. The objective is to minimize system cache costs while satisfying diverse QoS requirements for various vehicle user application requests. Firstly, a graph neural network model predicts cached content popularity using historical user request data. The SBS agents are modeled as graph vertices, with inter-agent relationships represented as edges, and the predicted cached content popularity integrated into the environment. The MARL framework further employs Graph Neural Networks (GNN) to extract effective inter-agent information, facilitating improved decision-making. Simulation results show that the proposed algorithm achieves faster convergence than baseline methods and effectively lowers cache costs while meeting diverse QoS requirements.

With the continuous expansion of the Low Earth Orbit (LEO) satellite scale, the formation of large LEO satellite constellations (LLSC) has significantly enhanced both network capacity and coverage capabilities. However, frequent handovers within multi-satellite coverage areas not only reduce handover efficiency but also adversely affect the quality of network services. This paper analyzes the relationship between constellation scale and system performance during the constellation design, aiming to minimize coverage time, handover frequency, and susceptibility to terrestrial interference. Under the condition of design parameters and seamless coverage requirements, a multi-objective optimization algorithm, namely the S-Metric Selection Evolutionary Multi-objective Optimization Algorithm (SMS-EMOA), is developed to optimize the LLSC configuration. Furthermore, an LLSC design scheme based on the Walker-delta and Walker-polar models is proposed, where both schemes have two different satellite service times, $\mu =300$ s and $\mu =600$ s. Simulation results demonstrate that compared to existing systems such as Kuiper and OneWeb, the designed Walker-delta LLSC achieves the minimum coverage time while exhibiting superior handover performance and stronger anti-interference ability.

With the proliferation of artificial intelligence applications, computing resources indicate a trend of ubiquitous deployment. However, traditional network architectures struggle to efficiently leverage computing resources for personalized request guarantees. Considering the unevenness and heterogeneity of computing resources, as well as the diversity of tasks in computing power networks (CPNs), SI-CRM, an social-aware integration of network connection, control, and service for computing resource management scheme is proposed. Based on social network analysis theory, this scheme models CPNs and comprehensively extracts endogenous social features from multiple dimensions. Then, propagation dynamics theory is utilized to predict the network cascade structure with the help of real-time sensing of node connectivity status, resource availability, and service requirements. Furthermore, endogenous social features are exploited to encourage resource node collaboration, and experience replay is introduced into deep reinforcement learning, thereby promoting proactive collaboration between connection and control in the network southbound. Finally, deep integration between control and service in the network northbound is realized by leveraging resource pre-scheduling and instance sharing. Simulation results verify the superiority of the proposed resource management scheme in reducing the overhead of CPNs.

Cyber-Physical Networks(CPN) are comprehensive systems that integrate information and physical domains, and are widely used in various fields such as online social networking, smart grids, and the Internet of Vehicles(IoV). With the increasing popularity of digital photography and Internet technology, more and more users are sharing images on CPN. However, many images are shared without any privacy processing, exposing hidden privacy risks and making sensitive content easily accessible to Artificial Intelligence(AI) algorithms. Existing image sharing methods lack fine-grained image sharing policies and cannot protect user privacy. To address this issue, we propose a social relationship-driven privacy customization protection model for publishers and co-photographers. We construct a heterogeneous social information network centered on social relationships, introduce a user intimacy evaluation method with time decay, and evaluate privacy levels considering user interest similarity. To protect user privacy while maintaining image appreciation, we design a lightweight face-swapping algorithm based on Generative Adversarial Network(GAN) to swap faces that need to be protected. Our proposed method minimizes the loss of image utility while satisfying privacy requirements, as shown by extensive theoretical and simulation analyses.