ABSTRACT The escalating deployment of Internet of Things (IoT) devices across critical infrastructure demands intrusion detection systems (IDS) capable of reliably discriminating hierarchically structured cyber threats such as denial‐of‐service, man‐in‐the‐middle, Mirai, and scan attacks. This study proposes a hybrid CNN‐GRU‐attention architecture in which parallel convolutional and recurrent branches are fused through a cross‐modal attention mechanism computing joint scores from spatial and temporal features via a shared scoring function, enabling adaptive per‐sample prioritisation of the more discriminative branch. ADASYN and ProWRAS oversampling strategies are compared within a leakage‐free pipeline across binary, category and sub‐category classification on IoTID20, and validated independently on TON‐IoT against six deep learning baselines. On IoTID20, test accuracies reach 1.0000, 0.9995 and 0.8228/0.8225 (ADASYN/ProWRAS). McNemar's tests show no significant difference between samplers, although per‐class analysis reveals a precision–recall trade‐off at the sub‐category tier. On TON‐IoT, the model achieves 0.9788 accuracy and F1‐score, exceeding six baselines on most metrics, trailing only in balanced accuracy due to one underrepresented class. Attention‐weight analysis shows concentration decreasing with granularity, with ProWRAS inducing a more distributed pattern that explains this trade‐off. An ablation study confirms attention and oversampling chiefly improve precision (0.8019–0.9005) over accuracy, though residual Mirai confusion reflects a feature‐space limitation.
ABSTRACT The rapid growth of Internet of things (IoT) deployments has intensified the need for effective and reliable intrusion detection systems capable of operating under heterogeneous and resource‐constrained environments. In response, deep learning (DL)‐based intrusion detection systems have been increasingly explored across IoT application domains. This study presents a PRISMA guided systematic review with descriptive meta‐analysis of recent DL‐based intrusion detection research in IoT environments, focusing on architectural trends, evaluation practices, deployment considerations and trust‐related properties. A total of 42 primary empirical studies published between 2021 and early 2026 were identified through systematic searches across major scholarly databases and analysed using standardized inclusion criteria and data extraction protocols. Instead of statistically aggregating heterogeneous results, the review employs descriptive meta‐analytic synthesis to examine patterns in model architectures, learning paradigms, deployment orientations and evaluation metrics. The analysis reveals a gradual shift from centralised DL models towards hybrid architectures, federated learning frameworks and resource‐aware designs suitable for edge or fog environments. Federated and privacy‐preserving learning approaches demonstrate promising detection performance while maintaining data locality, whereas explainable AI techniques are incorporated in only a limited subset of current IoT intrusion detection systems. The results also highlight a persistent gap between benchmark‐driven evaluation practices and the operational requirements of IoT deployments. Future studies should therefore adopt evaluation practices that jointly consider detection effectiveness, computational cost, privacy preservation, interpretability and deployment feasibility.
ABSTRACT Generative black‐box fuzzing techniques have been demonstrated to offer distinct advantages when confronted with closed‐source systems. However, testers have been shown to lack efficiency in developing fuzzing templates for emerging increasingly complex protocols. To address this challenge, we develop GeniFuzz and construct a proprietary knowledge database for generative fuzzing. This database is designed to refine and contextualise the large language models, thereby enabling it to substitute for the experts' function in generating fuzzing templates. Thus, a generative fuzzing framework based on large language models has been established. The experimental results demonstrate that GeniFuzz's composite score in the fuzzing template generation task is enhanced by 37.41% on average compared to the expert‐written templates, and by 22.48% over the expert‐written templates in terms of path coverage, and by 17.22% on average in efficiency of triggering crashes. Furthermore, the experiments demonstrate that the enhancement in the efficacy of fuzz testing, as implemented by the GeniFuzz framework, exhibits robustness and independence from the particular large language model utilised.
ABSTRACT This paper proposes a collision‐aware proximal policy optimisation (CA‐PPO) framework for adaptive resource management in grant‐free nonorthogonal multiple access (GF‐NOMA) enabled wireless federated learning systems. Unlike conventional approaches that rely on static clustering or grant‐based access, CA‐PPO dynamically coordinates device clustering and transmit‐power allocation under stochastic channel conditions, random‐access collisions and successive interference cancelation (SIC) constraints. The proposed framework employs a hybrid discrete–continuous actor–critic architecture with a collision‐aware reward that jointly captures decoding success, energy consumption, latency and retransmissions. CA‐PPO enables scalable and reliable aggregation in ultradense IoT deployments by improving participation stability and reducing communication overhead. Simulation results demonstrate that the proposed approach accelerates convergence by approximately 16%, reduces per‐device energy consumption by up to 25% and sustains high device participation under dense connectivity. These findings highlight the potential of learning‐driven network‐aware resource optimisation for next‐generation wireless federated learning systems.
ABSTRACT In modern computing and blockchain systems, with the growth of networks, classical consensus mechanisms, which tend to be hard and rule‐based, cannot dynamically adapt to changing workloads and, therefore, consume unnecessary energy and have uneven performance. The proposed AI–blockchain convergence framework is aimed at accomplishing energy‐efficient and self‐optimising computing networks based on reinforcement‐based consensus adaptation. The proposed model combines a multi‐agent reinforcement learning controller with blockchain telemetry which allows the real‐time adjustment of the block size, committee composition, and timeouts. Compared to the traditional consensus system, AIBLOCK adapts to both throughput and stability based on feedback of the live states on a network, inherently turning the blockchain into a self‐regulating digital organism. Large scale node deployments have been evaluated through experimentation and have a 34% lower energy use as well as a 22% increase in consensus stability when compared to HiCoOB and Layer‐2 baselines. The framework reported sublinear scaling of energy per node, and it converges quickly when subjected to dynamic loads in terms of transactions, which validate its adaptive efficiency. AIBLOCK, therefore, represents a transition to cognitive blockchain ecosystems that do not necessarily have to be manually tuned but learn instead.
This work explores the training mode of the physical culture literacy of ordinary college students based on big data analysis from the perspective of healthy China. The cultivation of sports cultural literacy of college students in Jiangsu Province is selected as the research object, and information on 900 randomly selected college students is collected by means of questionnaires and interviews. The results show that 56.35% of college students think that sports are an exercise, and 43.65% think that it is an education; 33.88% and 28.94% think that sports have cultural and political values, respectively; 50.84%, 60.94%, and 41.06% of college students know about repetitive, continuous, and circular exercise methods, respectively; and 68.35% score 60 ‐ 85 points in sports performance. The proportion of people who think that sports are or are essentially indispensable is 65.88%; the proportion of students taking regular exercise is 37.06%; the proportion of students with sports consumption higher than 1,000 is 6.47%; the proportion of students taking 1 ‐ 2 days of exercise per week is 60.71%; and 45.06% are more interested in physical education.. This article is protected by copyright. All rights reserved.
In recent years, cloud computing has emerged as a major player in the field of information knowledge. Cloud companies are building data centers (DCs) around the world to fulfill the increased demand for computing and storage resources. When it comes to cloud computing, load balancing refers to the practice of spreading workloads and computational properties among multiple nodes in a cluster. Organization can manage their workloads and applications better by distributing resources among PCs, networks, and servers. The AEHO (Adaptive Elephant Herd Optimization) algorithm was used in this study to design a load‐balancing system. The elephant population is alienated into clans, with each point on the elephant's trunk denoting a different type of solution. The best solution can be found in a family with a matriarch at the helm. For making and responding to tasks, the suggested AEHO algorithm was found to have an effective average load. Compared to other approaches, the simulation results demonstrated that the proposed AEHO model performed effectively. With an average response time of 13.58 milliseconds, a turnaround time of 21.09 milliseconds, reliability of 87%, and a throughput of 99 percent with a 45‐millisecond makespan for minor operations, the simulation results show that the AEHO algorithm outperformed all other compared approaches. This article is protected by copyright. All rights reserved.
Terahertz (THz) communication is a key enabler for 6G wireless networks but suffers from severe path loss and dynamic blockage, making conventional MAC protocols inefficient. This paper proposes a reconfigurable intelligent surface–based deep reinforcement learning MAC (RIS-DRL-MAC) framework that enables cross-layer optimisation between the physical and MAC layers. By embedding RIS perception features—such as equivalent channel gain and link stability—into the state space and jointly optimising beam direction, channel access and RIS phase configuration through a distributed twin delayed deep deterministic policy gradient (TD3) algorithm, the protocol achieves adaptive environment control. Simulation results show that, under dynamic blockage and high-load conditions, RIS-DRL-MAC improves network throughput by up to 90%, reduces access delay by 50% and maintains over 90% link availability compared with baseline schemes. The proposed method establishes a closed loop of sensing, decision and environment reconfiguration, providing an effective solution for reliable and energy-efficient THz mesh networking.
This paper presents an integrated optimisation framework for wireless sensor networks (WSNs) designed to manage the competing demands of energy efficiency, latency reduction, throughput improvement and communication reliability under dynamic and large-scale deployment conditions. The framework reorganises and enhances three core optimisation methods—genetic algorithm (GA), particle swarm optimisation (PSO) and an improved NSGA-II—by embedding adaptive behaviours and topology-aware decision logic. The GA is strengthened through a zone-oriented crossover mechanism and a sink-distribution-based initialisation strategy, which enhance coverage robustness and fault tolerance. The PSO module applies self-adjusting learning coefficients and QoS-aware routing constraints to maintain efficient path selection under varying load conditions. The improved NSGA-II incorporates an adaptive selection mechanism and a direction-guided crossover operator to better balance energy consumption and delay in multi-objective optimisation. Simulation results show that the proposed framework consistently outperforms federated DDQN and adaptive MOPSO across all performance indicators. It also demonstrates superior multi-objective convergence quality, achieving an IGD of 0.03 and an HV of 0.87. Overall, the framework enhances the scalability, resilience and operational efficiency of WSNs and provides practical guidance for adaptive scheduling in complex real-world environments.
Recently, precision agriculture has used wireless sensor networks (WSNs) to gain valuable insights and improve crop yields, promoting efficient resource use and data-driven decisions. However, WSNs face challenges, such as high power consumption from continuous sensing, data processing and communication, especially in large-scale setups, which limits their lifespan. This paper focuses on reducing power use in agricultural WSN sensor nodes during data transmission of soil moisture, rainfall, light intensity, air temperature and humidity from the transmitting sensor node to the base station. Four algorithms are proposed to cut power consumption. First, a sleep/wake (S/W) scheme using a simple duty cycle called S/W-DC. Second, the S/W scheme combined with adaptive data sampling (ADS) based on redundant data (RD), called S/W-ADS-RD. Third, the S/W scheme integrated with dynamic voltage scaling (DVS), named S/W-DVS. Fourth, a hybrid of all three, called S/W-ADS-RD-DVS. The sensor uses a 12 V/5 W solar panel for energy harvesting to maintain operation. The hybrid algorithm achieved 99.232% power savings and extended battery life to approximately 1.83 years. During a 6-h session, data transmission was reduced by 99.93%. This research could significantly improve WSN efficiency in precision agriculture and can be applied to energy-efficient WSN deployment across various fields, supporting Internet of Things (IoT) applications.
This work considers the use of optical wireless communications (OWC) for transmitting data from medical devices in wireless body-area networks (WBANs) for the purpose of patient vital sign monitoring. In such networks, the design of efficient medium-access control (MAC) protocols is crucial to ensuring reliable and effective data transmission from multiple nodes. Here, IEEE 802.15.6 and IEEE 802.15.7 standards, developed for wireless personal area networks (WPANs), are compared and evaluated through numerical simulations to assess their suitability for the specific use-case under consideration. The former standard was initially developed for radio-frequency (RF) networks, whereas the latter is based on OWC technology. This work also provides insights into the performance of the recently-introduced IEEE 802.15.13 standard, designed for optical WPANs. Our study relies on the Castalia simulator, combined with realistic optical WBAN channel models developed in our team's previous works, with network energy efficiency and quality-of-service (QoS) serving as the primary evaluation criteria. Both cases of intra- and extra-WBAN connectivity are considered, where the former refers to data transmission from medical sensors to a coordinator node (CN), and the latter to transmission from CNs (each corresponding to a patient) to an access point (AP), in a hospital ward, for instance. Additionally, two scenarios are examined: battery-operated CNs and power-outlet-connected CNs, with the latter assumed to be positioned on the patient's beds in an intensive care unit (ICU) room. Our results show the advantage of the IEEE 802.15.7 MAC protocol in terms of both energy consumption and QoS, for all considered scenarios. Finally, because the number of patients may vary across hospital wards, the scalability of the aforementioned MAC protocols is also investigated by varying the number of patients up to 8. The results indicate that IEEE 802.15.13, which relies on time-division multiple access (TDMA), is a viable candidate for optical WBANs despite its limited scalability, which could be resolved using a more flexible allocation of time resources to ensure that all nodes are granted access to the transmission time slots. Overall, this study advances current knowledge and offers new insights into the design of robust optical WBANs that can ensure acceptable QoS under varying conditions while preserving energy efficiency, enabling their practical deployment in real-world healthcare scenarios.
We study how far a diffusion process on a graph can deviate from a designed starting pattern when the pattern is generated via Laplacian regularisation. Under standard stability conditions for undirected, entrywise nonnegative graphs, we give a closed-form, instance-specific upper bound on the steady-state spread, measured as the relative change between the final and initial profiles. The bound separates two effects: (i) an irreducible term determined by the graph's maximum node degree, and (ii) a design-controlled term that shrinks as the regularisation strength increases (with an inverse square-root law). This leads to a design rule: given any target limit on spread, one can choose a sufficient regularisation strength in closed form. Although one motivating application is array beamforming – where the initial pattern is the squared magnitude of the beamformer weights – the result applies to any scenario that first enforces Laplacian smoothness and then evolves by linear diffusion on a graph. Overall, the guarantee is non-asymptotic, easy to compute, and certifies the maximum steady-state deviation.
The transformation of industrial systems through real‐time analytics, autonomous control, and intelligent data acquisition has underscored the pivotal role of Industrial Edge Computing (IEC) in next‐generation Industrial Internet of Things (IIoT) environments. By enabling decentralized processing close to data sources, IEC enhances responsiveness, supports latency‐sensitive applications, and reduces the strain on centralized infrastructure. However, as IIoT ecosystems grow in scale and complexity, traditional edge solutions face increasing challenges related to device heterogeneity, dynamic network conditions, bandwidth constraints, and energy efficiency. This special issue explores emerging advancements in edge intelligence that address these pressing challenges. It brings together innovative research that leverages artificial intelligence, advanced communication technologies, and architectural innovations to improve the adaptability, resilience, and performance of edge‐enabled industrial systems. The featured studies contribute novel techniques for real‐time data processing, secure and efficient communication, and intelligent decision‐making at the edge, all of which are essential for supporting industrial automation, predictive maintenance, and cyber‐physical operations. Collectively, these contributions highlight the immense potential of edge intelligence to redefine the operational landscape of industrial systems. This issue is intended to support ongoing research and practical innovation in the evolving domain of edge‐enabled IIoT technologies.
Coping with the unprecedented surge in traffic volume necessitates a profound overhaul of traditional networking architectures. In response, software-defined networking (SDN) has emerged as a groundbreaking architecture that separates the control plane from the data plane, relocating it to a more computationally capable central controller. This paradigm shift paves the way for integrating recent advancements in reinforcement learning (RL) for traffic engineering and routing. This paper presents a systematic guide to implementing this integration in Java-based, open-source, open-network operating system (ONOS) SDN controllers. The control plane implementation in ONOS and data plane implementation in Mininet constitute a holistic SDN framework for evaluating the performance of RL-based traffic engineering and routing schemes. Furthermore, we implement a direct-policy transfer algorithm to enhance the RL agent's reaction time to link failures in the network topology. Considering end-to-end delay, throughput, and packet-loss ratio as our performance evaluation metrics, we compare and contrast the performance of four existing schemes.
This study experiments with machine learning algorithms for detecting distributed denial of service attacks as a multiclass classification problem. The algorithms included the K-nearest neighbours, decision trees, support vector machines, random forests, extreme gradient boosting, gradient boosting machines and multilayer perceptron. We validated the models using the hold-out and cross-validation methods, performed class and model ablation analysis to evaluate performance impacts and applied feature selection techniques, feature importance and statistical tests. For instance, using 10-fold cross-validation with 79 features, 11 attack types and regular network traffic, the tree-based models achieved accuracies ranging from 75.69% to 76.24%. When using 15 features, seven attacks and regular network traffic, model accuracy improved significantly, ranging from 97.77% to 98.08%. Furthermore, in specific application scenarios, some models achieved near-perfect classification performance. Decision tree achieved the highest accuracy score for the local network communication scenario, reaching 99.86%, followed by software distribution or updates at 99.70%, web platforms and online applications at 98.25%, video streaming or online gaming at 97.06%, infrastructure monitoring and management at 95.00% and directory services and corporate authentication at 87.15%. Depending on the application scenario, our results indicate that specialised models can support classification tasks targeting specific system components with high performance.
The deployment of unmanned aerial vehicles (UAVs) as aerial base stations in cellular networks presents a dynamic solution to meet the demands of high and fluctuating traffic patterns. Efficient placement of UAVs is crucial to harness their benefits and adapt intelligently to environmental changes. This paper introduces a multi-objective optimisation model aimed at maximising user coverage and minimising overlap among drone-based base stations in 6G networks. To address this optimisation issue, the Nondominated Sorting Genetic Algorithm II (NSGA-II) is deployed, enabling the identification of Pareto optimal solutions that strike a balance between conflicting objectives. Through simulations conducted under various scenarios, the proposed model demonstrated significant improvements in user coverage and reduction of overlap among base stations compared to existing techniques. The findings reveal the effectiveness of the proposed model in balancing the objectives of coverage and overlap, resulting in an enhanced 6G network design. The method achieves an average coverage probability of 98.39% and an average overlap improvement percentage (OIP) of 92.39%, validated through 50 experimental runs. These results underscore the robustness and superiority of the proposed NSGA-II-based strategy in optimising DBS placement, contributing to the advancement of 6G cellular networks.
Vehicular ad hoc networks (VANETs) are an essential enabler of intelligent transport systems (ITS), facilitating real-time communication among vehicles to enhance traffic safety and mobility. However, challenges such as high node mobility, frequent topology changes, and variable network density continue to impede the design of reliable and efficient routing protocols. This paper proposes CARAC (Cluster-based Ant-colony Routing with Adaptive Cluster Head), a hybrid routing protocol that integrates dynamic clustering using the K-medoids algorithm with Ant Colony Optimisation (ACO) to improve route stability, scalability, and data delivery performance. CARAC forms mobility-aware clusters by grouping vehicles based on spatial proximity and relative velocity. Each node periodically computes a local stability index to evaluate its membership within a cluster, allowing adaptive cluster maintenance. The protocol also incorporates ACO for optimal path selection and utilises Road Side Units (RSUs) as relays when direct communication between cluster heads is not feasible. Simulation experiments conducted in NS-3 with a vehicular network scenario demonstrate that CARAC consistently outperforms benchmark protocols such as AQRV, MetaLearn and CPB. It delivers higher route discovery success, better packet delivery performance, lower latency, and greater throughput. These results validate the advantages of combining clustering, bio-inspired optimisation, and adaptive stability evaluation in VANET routing, positioning CARAC as a robust and scalable solution for next-generation ITS applications.
Large scale software-defined networks have two main concerns, which are scalability and reliability. One of the problems with multi-controller architectures in these networks is the static mapping between SDN switches and controllers, which prevents the control plane from adapting to traffic changes. The dynamic mapping between switches and controllers by migrating switches from highly loaded controllers to lightly loaded controllers can provide compatibility for the control plane. This paper first discusses multi-controller load balancing as a key research challenge and then explores switch migration as a solution. A variety of load balancing methods based on switch migration are investigated and then a comprehensive comparison is made between them.
The fast evolution of cyberattacks in the Internet of Things (IoT) area, presents new security challenges concerning Zero Day (ZD) attacks, due to the growth of both numbers and the diversity of new cyberattacks. Furthermore, Intrusion Detection System (IDSs) relying on a dataset of historical or signature-based datasets often perform poorly in ZD detection. A new technique for detecting zero-day (ZD) attacks in IoT-based Conventional Spiking Neural Networks (CSNN), termed ZD-CSNN, is proposed. The model comprises three key levels: (1) Data Pre-processing, in this level a thorough cleaning process is applied to the CIC IoT Dataset 2023, which contains both malicious and the most recent attack patterns in network traffic, ensuring data quality for analysis, (2) CSNN-based Detection, where outlier identification is conducted by comparing two dataset groups (the normal set and the attack set) within the same time period to enhance anomaly detection and (3) In the evaluation level, the detection performance of the proposed model is assessed by comparing it with two benchmark models: ZD-Deep Learning (ZD-DL) and ZD- Convolutional Neural Network (ZD-CNN). The implementation results demonstrate that ZD- CSNN achieves superior accuracy in detecting zero-day attacks compared to both ZD-DL and ZD-CNN.
This paper investigates a vulnerability in IEEE 802.11 wireless local area networks, focusing on a MAC sublayer attack known as acknowledgement (Ack) spoofing. The paper delves into the distributed coordination function (DCF) and examines how Ack spoofing attacks affect network performance by manipulating the Ack operation essential for successful data exchange between stations. This manipulation disrupts Ack-based rate control mechanisms and the backoff procedure of the standard, leading to decreased performance for legitimate receivers with lossy links to access points. The paper introduces strategies to perform Ack spoofing attacks. To counter these threats, a novel detection and mitigation technique is proposed that effectively detects and mitigates any of the proposed attack strategies. The introduced technique is simple to implement, compatible with all versions of the existing IEEE 802.11 standard and all rate control mechanisms that rely on Ack frames in their operations. It also requires no modifications to the existing IEEE 802.11 standard, facilitating easy adoption by manufacturers. Moreover, it leverages a unique approach that avoids a cross-layer design, maintaining the integrity of layer abstraction. Through detailed simulations and analysis, the effectiveness of the proposed attack strategies and the detection and mitigation technique is demonstrated under various scenarios.
