TNSM Searchable Index

Author(s)	Title	Year	Publication	Keywords
Guangxia Xu, Zhuo Ye, Lu Wang, Xing Huang, Lei Liu, Shahid Mumtaz, Mohsen Guizani	GNN-OSS: A Capacity-Feasible Graph Learning Framework for Secure Blockchain Sharding in IIoT	2026	Early Access	Graph neural networks Sharding Information rates	Effective scaling of blockchain-enabled Industrial Internet of Things (IIoT) requires sharding that simultaneously ensures transaction locality, strict committee-size feasibility, and robustness against malicious node concentration. Existing methods often fail to balance this trilemma, risking either infeasible deployments or increased shard-takeover vulnerabilities. To address this, we propose GNN-OSS, a deployable sharding framework that decouples topology-aware preference learning from hard constraint enforcement. It first employs a trust-repulsion graph neural network to learn locality-aware preferences while discouraging low-trust nodes from collapsing into the same representation region. A Post-Hoc Capacity-Constrained Projection (PH-CCP) then maps these soft preferences into strictly feasible shard assignments. Finally, an entropy-driven Over-lapping Sparse Scheme (OSS) selectively replicates boundary nodes to reduce residual cross-shard overhead without altering primary consensus membership. Evaluations demonstrate that, under the evaluated settings, GNN-OSS achieves a favorable performance–security trade-off. Against 20% malicious nodes, it substantially mitigates shard-takeover risks. Furthermore, it improves throughput by up to 33% over strictly feasible baselines and lowers the cross-shard ratio from 6.4% to 4.4% with minimal per-epoch overhead. Overall, GNN-OSS provides a practical sharding framework for open or hybrid blockchain-enabled IIoT environments.	10.1109/TNSM.2026.3709024
Domenico Tortola, Mauro Orrù, Paolo Mori, Laura Ricci	E-CUBE: A Cross-Chain Erasmus+ Management System	2026	Early Access		This paper presents E-CUBE (Erasmus+ through Cosmos Unified Blockchain Environment), a decentralized application for managing the student mobility in the context of the Erasmus+ program using blockchain technology and crosschain solutions. Built on the Cosmos framework and its Inter- Blockchain Communication protocol, E-CUBE enables seamless interoperability between independent blockchain networks operated by national university consortia for Erasmus+ program management. Each university consortium runs its own customized blockchain, reflecting country-specific regulations, while a shared hub chain facilitates secure and verifiable cross-chain exchanges. The system supports and automates critical Erasmus+ processes, such as student registration, Erasmus+ application submission, Learning Agreement management, and credit transfer— in order to reduce manual workload and ensure consistency across institutions. E-CUBE’s architecture also supports dynamic operations like extending or ending a mobility period early, with all data synchronized across chains via cross-chain exchanges. A working prototype was developed and deployed in a distributed environment, obtained by deploying a total of 10 nodes across the entire European territory, in order to simulate realistic cross-border mobility scenarios. Experimental evaluations show that the system maintains low latency, and incurs minimal gas fees per operation, confirming its technical feasibility and efficiency for real-world deployment in higher education.	10.1109/TNSM.2026.3710003
Deemah H. Tashman, Soumaya Cherkaoui	Trustworthy AI-Driven Dynamic Hybrid RIS: Joint Optimization and Reward Poisoning-Resilient Control in Cognitive MISO Networks	2026	Early Access	Reconfigurable intelligent surfaces Reliability Optimization	Cognitive radio networks (CRNs) are a key mechanism for alleviating spectrum scarcity by enabling secondary users (SUs) to opportunistically access licensed frequency bands without harmful interference to primary users (PUs). To address unreliable direct SU links and energy constraints common in next-generation wireless networks, this work introduces an adaptive, energy-aware hybrid reconfigurable intelligent surface (RIS) for underlay multiple-input single-output (MISO) CRNs. Distinct from prior approaches relying on static RIS architectures, our proposed RIS dynamically alternates between passive and active operation modes in real time according to harvested energy availability. We also model our scenario under practical hardware impairments and cascaded fading channels. We formulate and solve a joint transmit beamforming and RIS phase optimization problem via the soft actor-critic (SAC) deep reinforcement learning (DRL) method, leveraging its robustness in continuous and highly dynamic environments. Notably, we conduct the first systematic study of reward poisoning attacks on DRL agents in RIS-enhanced CRNs, and propose a lightweight, real-time defense based on reward clipping and statistical anomaly filtering. Numerical results demonstrate that the SAC-based approach consistently outperforms established DRL base-lines, and that the dynamic hybrid RIS strikes a superior trade-off between throughput and energy consumption compared to fully passive and fully active alternatives. We further show the effectiveness of our defense in maintaining SU performance even under adversarial conditions. Our results advance the practical and secure deployment of RIS-assisted CRNs, and highlight crucial design insights for energy-constrained wireless systems.	10.1109/TNSM.2026.3660728
Liang Chen, Xiaoding Wang, Limei Lin, Yanze Huang, Siwei Zheng	Defense in Depth: Architectural Homology for Adversarially Robust Semantic Communication	2026	Early Access		Semantic communication framework based on deep encoder–decoder architectures are increasingly vulnerable to adversarial attacks, wherein imperceptible input perturbations can induce significant misclassifications, posing critical risks to next-generation communication networks. To address this challenge, we introduce TopAliSC-KG, a holistic robust learning framework designed to fortify semantic communication against adversarial threats through a multi-layered defense strategy. Our approach integrates two complementary mechanisms: a homology graph-aligned adversarial training module that embeds topological constraints into the model optimization process to promote structural invariance, and a knowledge-guided semantic consistency module that utilizes auxiliary models to inject stable, high-level semantic information into the training loop. Together, these components establish a defense-in-depth architecture that enhances resilience across data, feature, and model levels. Extensive evaluations across diverse channel conditions, signal-to-noise ratios, and adversarial attack scenarios show that TopAliSC-KG consistently improves adversarial accuracy by 0.32%–3.39%, with the knowledge guidance mechanism contributing a further 0.11%–0.21% gain. This work provides a validated, multi-framework defense strategy suitable for securing semantic communication in mission-critical applications, advancing both the security and reliability of intelligent communication systems.	10.1109/TNSM.2026.3705940
Manish Bansal, J. Harshan	Low-Latency Spatial-Provenance Recovery Methods for Privacy-Constrained Vehicular Networks	2026	Early Access		In multihop Vehicle-to-Everything (V2X) networks, Road Side Units (RSUs) intend to collect information on vehicles’ location in a low-latency manner while respecting their privacy constraints to support real-time location-based services. To facilitate data collection, provenance is known to ensure trust and accountability of data. Although existing joint data- and spatial-provenance techniques preserve the privacy of vehicles up to a certain granularity with respect to the RSU and other vehicles, they are unsuitable when stringent deadlines are imposed on the end-to-end delay on the packets. As a consequence, there is a need for designing spatial-provenance methods for V2X networks that satisfy stringent deadlines on the end-to-end delays while managing the privacy concerns. To fill this research gap, we propose two novel protocols, namely: Bi-Segment Embedding (BSE) and Tri-Segment Embedding (TSE), which provide a skipping mechanism for joint data- and spatial-provenance while trading off privacy features among the vehicles. Through an extensive theoretical framework, we provide an analysis of the proposed schemes in terms of reliability, privacy, and communication overhead. When compared to the baselines, our protocols offer lower end-to-end delay, higher reliability in provenance reconstruction, and the same level of privacy with respect to the RSU. We validate latency gains using practical radio parameters, and our study reveals that our proposed protocols offer significant benefits in latency when implemented over a 5G stack.	10.1109/TNSM.2026.3710000
Lion Steger, Liming Kuang, Johannes Zirngibl, Georg Carle, Oliver Gasser	Still on Target? An Evaluation of IPv6 Target Generation Algorithms	2026	Early Access		Internet measurements are a crucial foundation of IPv6-related research. Due to the infeasibility of full address space scans for IPv6 however, those measurements rely on collections of reliably responsive, unbiased addresses, as provided e.g., by the IPv6 Hitlist service. Although used for various use cases, the hitlist provides an unfiltered list of responsive addresses, the hosts behind which can come from a range of different networks and devices, such as web servers, customer-premises equipment (CPE) devices, and Internet infrastructure. In this paper, we demonstrate the importance of tailoring hitlists in accordance with the research goal in question. By using PeeringDB we classify hitlist addresses into six different network categories, uncovering that 42% of hitlist addresses are in ISP networks. Moreover, we show the different behavior of those addresses depending on their respective category, e.g., ISP addresses exhibiting a relatively low lifetime. Furthermore, we analyze different Target Generation Algorithms (TGAs), which are used to increase the coverage of IPv6 measurements by generating new responsive targets for scans. We use seed sets, e.g., based on the categorized Hitlist. We evaluate the performance of TGAs under various conditions and find generated addresses to show vastly differing responsiveness levels for different TGAs. Furthermore, we evaluate of algorithm run times and differences between multiple TGA runs.	10.1109/TNSM.2026.3705935
Emilio Paolini, Andrea Pinto, Luca Valcarenghi, Flavio Esposito	Programmable In-Network Aggregation for Communication-Aware Federated Learning in 5G RANs	2026	Early Access	Modeling Timing Training	Federated Learning (FL) enables collaborative model training without sharing raw data, making it attractive for privacy-preserving applications at the wireless edge. However, when executed over real 5G networks, FL performance degrades due to uplink congestion, heterogeneous client capabilities, and intermittent connectivity. Most existing approaches attempt to mitigate these issues indirectly by optimizing clients (through adaptive participation, local training, or selection strategies) or by optimizing models (via pruning, quantization, or compression), but they ignore potential network bottlenecks. This paper introduces FLAG, an FL architecture that embeds innetwork aggregation directly into 5G gNodeBs, transforming the network into an active participant in the learning process. In particular, FLAG performs parameter aggregation at line rate within the 5G Service Data Adaptation Protocol layer and incorporates three mechanisms: Partial-Contribution Correction for loss-tolerant averaging, a timer-driven pipeline for real-time scheduling, and a deadline-based grouping strategy to mitigate stragglers. Experiments with realistic wireless emulation show that FLAG achieves up to 5.1× faster time-to-accuracy and maintains accuracy within 0.8% of a loss-free baseline, while reducing gNB-to-server bandwidth by aggregating pergNB rather than per-client. FLAG requires no modifications to clients or the parameter server, demonstrating how 5G-aware system design can make federated learning scalable, efficient, and resilient under real-world wireless conditions.	10.1109/TNSM.2026.3697723
Huijuan Zhu, Chenhao Zheng, Zhongyuan Liu, Yuan Zhang	Reliable Interpretations of Deep Learning-based Malware Detectors via Deep Q-Networks	2026	Early Access	Malware Signal detection Modeling	Deep learning has become widely used in Android malware detection, but its black-box nature raises trust concerns, limiting its use in critical security areas. To address this, various interpretation methods have been proposed. Unfortunately, these solutions often suffer from inconsistent results and poor adaptability to model updates. In this work, we propose XDQNMal, a Deep Q-Networks (DQN)-based global interpretation framework designed to uncover the critical features that drive decisions in deep learning-based malware detectors. To enhance the reliability of interpretation, XDQNMal captures API call frequency features derived from the runtime behavior of each application (App). Then, it unites a DQN model with the TabPFN detection model to work collaboratively, using variations in detection results as reward signals. These signals guide the DQN model to gradually identify the most impactful features as interpretations for the detection model’s decisions. Our experimental evaluation on real-world datasets demonstrates that the proposed XDQNMal framework generates reliable interpretation for deep learning-based malware detection models. For instance, suppressing the critical features identified by XDQNMal leads to an average decrease of 20.30% in the probability that the malicious sample is predicted as malicious, highlighting the pivotal role these features play in the model’s decision-making.	10.1109/TNSM.2026.3699408
Kunpeng Zheng, Huibin Zhang, Yongli Zhao, Yuan Cao, Wei Wang, Xin Li, Zhuangzhuang Ma, Lihan Zhao, Jie Zhang	Sun-Outage-Aware Topology Modeling and Adaptive Routing for Optical Satellite Networks	2026	Early Access	Sun Interrupters Joining processes	Optical satellite networks, supported by optical inter-satellite links (OISLs), provide reliable and low-latency optical connectivity. However, periodic and predictable sun outage events significantly compromise OISL availability, leading to frequent OISL interruptions and reduced network reliability. Existing routing algorithms often overlook the regularity of sun outage-induced interrupts and their differentiated impacts on services, resulting in degraded service performance. To address this challenge, this paper proposes a sun outage-enhanced time discretization OISL model and introduces a sun outage link-aware routing (SOLR) algorithm. By incorporating joint awareness of sun outage patterns and service requirements, SOLR employs an adaptive optimization mechanism to dynamically adjust routing decisions within temporal windows. Experimental results demonstrate that SOLR extends stable path durations by 39.9%, reduces interruption rates by 28.5%, and decreases blocking rates by 36.4%, significantly outperforming link-state-based routing algorithms. By effectively mitigating the impact of sun outages, SOLR ensures continuous optical service connections. This interruption-tolerant framework bridges network modeling and service provisioning, offering a robust solution for mission-critical service in optical satellite networks.	10.1109/TNSM.2026.3697856
Yiyang Li, Wei Wang, Yibo Wang, Qiaojun Hu, Weiliang Zhang, Yongli Zhao, Xiaoyu Wang, Jie Zhang	Computing-State Driven Proactive Congestion Control for AI Cluster Interconnect Networks	2026	Early Access	Timing Modeling Fluid flow	The rapid upgrade of computing power and the prosperity of large language model (LLM) in data center networks (DCNs) lead to a rigorous demand for ultra-low latency and high throughput. To mitigate the overhead of collective communication during distributed training (DT), Remote Direct Memory Access (RDMA) has been widely adopted in DCNs. Particularly, congestion control algorithms (CCAs) designed for RDMA have attracted much attention to mitigate performance deterioration under network congestion. However, through comprehensive analysis, we investigate that, due to sluggish end-to-end reaction and slow rate convergence, existing widely used reactive CCAs have several limitations in handling bursty traffic (e.g., AllReduce). Specifically, excessive packets are transmitted before senders activate the reaction and converge to the fair rate, which builds up a deep queue and may incur subsequent significant throughput loss. In this paper, we propose a computing-state driven proactive congestion control (CSPCC) with easy deployability. CSPCC consists of the congestion prediction module and the active congestion response module. It leverages current computing state to predict network congestion time and inform corresponding sources in advance. We provide a detailed introduction to the implementation of CSPCC. Then, we conducted small-scale hardware tests and large-scale simulations to evaluate the performance of CSPCC. On our testbed, under NCCL-TESTs, CSPCC improves throughput by 1.67%–13.35% and decreases switch queue occupancy by 28.33%–58.33% compared to DCQCN. Furthermore, under concurrent multi-job LLaMA training, it reduces end-to-end job completion time (JCT) by 5.3%–9.0%.	10.1109/TNSM.2026.3705429
Heewon Kim, Hochan Lee, Chanbin Bae, Haneul Ko, Sangheon Pack	Traffic- and Multi-Tenancy-Aware In-Network Aggregation Placement for Distributed Machine Learning	2026	Early Access	Memory Magnesium Modeling	Distributed machine learning is an effective method to alleviate the intensive computation costs of training; however, it suffers from network bottlenecks while collecting local results. The recent advent of programmable data planes has opened a new avenue, in-network aggregation, which executes gradient aggregations in the middle of the network, resolving network bottlenecks, and further accelerates distributed machine learning. However, due to resource-constrained features of current programmable data planes, deploying in-network aggregation functionalities throughout the network would impose an unacceptable burden, posing a need for sophisticated deployment. In this paper, a problem of deploying in-network aggregation functionalities is studied to minimize the total network traffic in multi-tenant distributed machine learning. We formulate the problem as an integer linear programming (ILP) problem and prove its NP-hardness. Since finding the optimal solution using the brute-force method is extremely complicated, we propose a traffic-aware in-network aggregation placement algorithm based on a two-stage many-to-one matching game (denoted TAPINA-MG). The simulation results demonstrate that TAPINA-MG shows nearoptimal performance with low complexity, achieving up to 22.5%, 38.9%, and 96.0% reduction for network traffic, maximum link utilization, and communication time, respectively, compared to state of the art, and effectively handles dynamic situations with minimal migration delay and comparable traffic performance.	10.1109/TNSM.2026.3709103
Juan Zhang, Yangjun Ma, Xunzheng Zhang, Zhao Huang, Qiuji Yi, Nauman Aslam	Multi-objective SFC Placement with Future Demand Awareness in Dynamic Cross-Domain Networks	2026	Early Access	Modeling Optimization Transformers	Efficient service function chain (SFC) placement is critical for optimizing network service delivery in dynamic cross-domain networks (CDNs), especially under resource-constrained and heterogeneous environments. However, existing approaches face fundamental limitations in achieving effective multi-objective optimization, particularly in balancing latency minimization with efficient resource utilization. These challenges are further compounded by the inability to capture future resource dynamics and limited visibility across multiple domains. To address these challenges, we propose a novel multi-objective framework for SFC placement that jointly considers latency and resource utilization. The framework integrates Transformer-based prediction with linear programming (LP) to explicitly model future deployability, enabling proactive and globally informed placement decisions. In addition, a dynamic modeling mechanism is developed using domain-aware detection and graph autoencoders (GAEs) to capture evolving network topologies and cross-domain structural dependencies. A Pareto-based optimization strategy is further employed to systematically balance latency and resource efficiency across heterogeneous domains and varying workload conditions. Extensive experiments across multiple network scales and diverse SFC configurations demonstrate that the proposed framework achieves a superior trade-off between latency and deployment capability, while improving scalability, robustness, and long-term resource efficiency in dynamic and large-scale CDN environments.	10.1109/TNSM.2026.3708714
Yahuza Bello, Ahmed Refaey, Ping Yang	Secure Multi-Timescale Orchestration for Zero-Trust Cross-Datacenter Networks	2026	Early Access	Authentication Optimization Resource management	The widespread deployment of geographically distributed Data Centers (DCs) has intensified the need for scalable and secure access control mechanisms across Cross-Datacenter Networks (CDNs). Zero Trust Architecture (ZTA) addresses this need by enforcing continuous authentication and authorization through Policy Decision Points (PDPs); however, determining where to deploy PDPs and how to dynamically assign authentication requests in the CDNs remains a challenging and NP-hard problem. This challenge arises from the tight coupling between long-term placement decisions and short-term, stochastic authentication workloads. In this paper, we formulate a joint PDP placement and authentication assignment problem for zero-trust-enabled CDNs that minimizes deployment cost, authentication assignment cost, bandwidth consumption, and the number of active PDP instances under resource constraints. To efficiently solve the problem, we propose a Hybrid Hierarchical Deep Reinforcement Learning (HHDRL) framework that decomposes decision-making across multiple time scales. A high-level Double Deep Q-Network (DDQN) agent learns long-term PDP placement policies, while multiple low-level Asynchronous Advantage Actor–Critic (A3C) agents perform real-time authentication assignment within each DC. Extensive simulations demonstrate that the proposed DDQN–A3C framework converges reliably and consistently outperforms benchmark schemes, including DDQN–A2C, a single-agent DDQN approach, and a greedy baseline, achieving lower overall system cost and improved scalability with modest computational overhead.	10.1109/TNSM.2026.3707392
Madhura Adeppady, Yenchia Yu, Ali Rahmanian, Ahmed Ali-Eldin Hassan, Carla Fabiana Chiasserini	Efficient Management of Composite Heterogeneous Applications at the Network Edge	2026	Early Access	Central Processing Unit Servers Resource management	Edge computing is a promising paradigm for deploying latency-sensitive applications (Apps) as it brings resources closer to end users. Edge Apps often adopt a microservice (MS) architecture, breaking monolithic Apps into lightweight, containerized MSs that can be dynamically and independently deployed. However, managing such Apps involves three key challenges: (i) optimizing the placement of MSs to reduce both response time and resource overhead, (ii) handling MS migration or relocation as users move while minimizing App service disruption (App downtime), and (iii) enabling MS sharing across Apps while ensuring performance guarantees. We formulate this as an optimization problem, named Multi-microservice Application Placement (MAP), prove its NP-hardness, and introduce STEP (State and Topology-aware Edge-MS Placement), a polynomial-time heuristic. STEP distinguishes itself from prior work by: (i) jointly considering stateful and stateless MS characteristics in deployment decisions, (ii) exploiting MS shareability to reduce resource usage, (iii) balancing response latency, App downtime, and resource utilization, and (iv) leveraging multiple versions of the same MS to adapt quality of service to available edge resources. Our results in a small-scale scenario show that STEP achieves near-optimal performance with only 7% higher CPU cost than the optimal solution. Large-scale real-time experiments on a Kubernetes cluster demonstrate that STEP consistently outperforms competing methods, achieving up to 50% lower deployment costs while delivering 50% gain in app quality and saving 15% in radio resources with over 90% request success rates.	10.1109/TNSM.2026.3709656
Gergely Dobreff, Nóra Szlovencsák, Alija Pašić	A Framework for Disaster-Tolerant Slice Placement in Future Networks	2026	Early Access	Costing Costs Codes	Autonomous vehicles and telesurgery are placing increasing pressure on network operators to ensure that 5G and beyond networks can support a wide range of services with diverse and stringent requirements. Technologies such as Software-Defined Networking (SDN), Network Function Virtualization (NFV), and network slicing are key enablers for building an ecosystem capable of meeting these demanding conditions. Ensuring not only classical Quality of Service (QoS) metrics but also network resiliency is crucial, as failures in shared infrastructures can severely impact critical services. This paper addresses the problem of resilient network slice placement under arbitrary disasters or attacks, modeled as Shared Risk Link Group (SRLG) failure patterns. We propose an approach that guarantees strict end-to-end delay, bandwidth, and computing requirements while minimizing overall resource usage by accounting for potential failure scenarios. To this end, we introduce a Disaster-Tolerant Slice Placement Framework that enables network operators to define their own resilience scenarios and optimize the network accordingly. Several - routing and network coding–based - strategies are proposed and analyzed. We formulate the problem as an Integer Linear Program (ILP), analyze its computational complexity, and develop efficient heuristic algorithms to obtain near-optimal solutions. Extensive simulations demonstrate the effectiveness of the proposed methods in achieving resource-efficient and resilient network slice placement. The results show that high levels of resiliency can be achieved without excessive over-provisioning, positioning the proposed framework as an effective offline planning and benchmarking tool for 5G and beyond network design.	10.1109/TNSM.2026.3706661
Jing Zhang, Chao Luo, Rui Shao	MTG-GAN: A Masked Temporal Graph Generative Adversarial Network for Cross-Domain System Log Anomaly Detection	2026	Early Access	Anomaly detection Adaptation models Generative adversarial networks	Anomaly detection of system logs is crucial for the service management of large-scale information systems. Nowadays, log anomaly detection faces two main challenges: 1) capturing evolving temporal dependencies between log events to adaptively tackle with emerging anomaly patterns, 2) and maintaining high detection capabilities across varies data distributions. Existing methods rely heavily on domain-specific data features, making it challenging to handle the heterogeneity and temporal dynamics of log data. This limitation restricts the deployment of anomaly detection systems in practical environments. In this article, a novel framework, Masked Temporal Graph Generative Adversarial Network (MTG-GAN), is proposed for both conventional and cross-domain log anomaly detection. The model enhances the detection capability for emerging abnormal patterns in system log data by introducing an adaptive masking mechanism that combines generative adversarial networks with graph contrastive learning. Additionally, MTG-GAN reduces dependency on specific data distribution and improves model generalization by using diffused graph adjacency information deriving from temporal relevance of event sequence, which can be conducive to improve cross-domain detection performance. Experimental results demonstrate that MTG-GAN outperforms existing methods on multiple real-world datasets in both conventional and cross-domain log anomaly detection.	10.1109/TNSM.2026.3654642
Ashiqur Rahaman Ridoy, Arnab Kumar Biswas	Adaptive Intrusion Detection Systems: Leveraging Meta-Learning for Improved Cybersecurity	2026	Early Access	Modeling Fluid flow Labeling	In the evolving landscape of cybersecurity, the integration of machine learning (ML) into Intrusion Detection Systems (IDS) has become critical for detecting both known and unknown attacks. This paper proposes a novel multi-stage hybrid IDS framework combining unsupervised anomaly detection, supervised classification, and low-shot adaptation for enhanced resilience to concept drift. The architecture comprises three interconnected stages: Stage 1 (unsupervised anomaly gating) and Stage 2 (supervised taxonomy learning) operate in parallel on a shared harmonized feature space; Stage 3 (Hybrid Low-Shot Adapter (H-LSA)) performs low-shot adaptation when the Stage 1 trigger fires, using transferred Stage 2 weights and a prototype-based cosine-kNN jury. Within the meta-learning family, we instantiate a metric-based low-shot adaptation approach eschewing second-order Model-Agnostic Meta-Learning (MAML) in favor of a partial-freeze, first-order protocol with a prototype-based cosine-kNN jury to enable rapid, low-resource adaptation. Extensive experiments were conducted on the CICIDS2017 (Source), CSECIC-IDS2018 (Target), and the modern BCCC-cPacket-Cloud-DDoS-2024 (Target) datasets (hereafter referred to as BCCC-2024). The results demonstrate that while static Stage 2 models suffer catastrophic failure under concept drift (dropping to 45.36% and 38.32% accuracy on CICIDS2018 and harmonized BCCC-2024, respectively), the proposed framework successfully adapts to new environments, achieving 90.64% accuracy on CICIDS2018 (Macro-F1: 0.8981) and 89.70% on BCCC-2024 (Macro-F1: 0.8801) with a low-resource support set of only 500 labeled samples per class. Furthermore, the system exhibits high computational efficiency, achieving a Stage 3 adapted inference latency between 0.0786 ms and 0.1667 ms per flow across diverse traffic profiles, proving its suitability for real-time, scalable deployment in modern cloud and edge network infrastructures.	10.1109/TNSM.2026.3706597
Soonbeom Kwon, Yusu Noh, Youngwoo Jang, Illyoung Choi, Byungchul Tak, In-geol Chun, Young-Kyoon Suh	Scalable and Robust Resource Provisioning via Adaptive Task Scheduling for Edge Devices	2026	Early Access	Schedules Scheduling Cloning	Edge devices, such as wearables, drones, and CCTV systems, are vital for real-time data collection in urban intelligence. However, their limited computational and storage capacities pose significant challenges. While offloading to public clouds offers scalability, it often incurs high latency and operational costs. Conversely, centralizing workloads on edge servers may result in the underutilization of high-performance edge devices. To address these limitations, we introduce ERPF, a Kubernetes-based Edge Resource Provisioning Framework that augments the capabilities of heterogeneous edge environments. ERPF orchestrates dynamic volume provisioning, GPU-aware resource allocation, execution context migration, and adaptive task distribution to improve system flexibility and efficiency. Building on this, we propose a novel adaptive task scheduling technique, termed eATS, composed of three key mechanisms: (i) Partition Smoothing Scheme for stable task granularity control, (ii) Resilient Edge Reintegration for failure detection and task reassignment, and (iii) Competitive Task Cloning for speculative execution with fastest-result commitment. The proposed eATS scheme reduces task execution time by up to 27.6%, lowers partition size variability by 8.7×, and improves scheduling robustness across heterogeneous edge devices over the baseline.	10.1109/TNSM.2026.3694238
Arash Heidari, Jamal N. Al-Karaki	NOVA: A Self-Supervised Graph Framework for Real-Time Anomaly Detection in Internet of Vehicles	2026	Early Access	Context Internet of Vehicles Modeling	The Internet of Vehicles (IoV) enables cooperative driving and real-time Vehicle-to-Everything (V2X) communication but remains vulnerable to behavioral and structural anomalies due to its dynamic, decentralized nature. Existing deep learning methods either overlook topological inconsistencies or ignore communication feature fidelity, while random-walk sampling introduces contextual noise. In this paper, we propose Network Observation for Vehicular Anomalies (NOVA), a self-supervised graph-based framework that detects both behavioral and structural anomalies in IoV networks without labeled data. NOVA models vehicular communications as attributed graphs and employs intimacy-guided subgraph sampling to extract meaningful neighborhoods. A Graph Convolutional Network (GCN)–based generative module reconstructs node attributes to reveal behavioral deviations, while a contrastive module validates structural coherence through embedding comparisons of real and perturbed contexts. Their hybrid anomaly score enables accurate, scalable, and real-time detection of compromised nodes. Performance results show that NOVA achieves state-of-the-art performance (98.7% accuracy, 98.1% F1), real-time throughput (~4.7k events/s at 5k msg/s), and strong robustness (AUROC 0.99, AUPRC 0.98, FAR 0.05) with near-linear scalability (≤40 ms latency for 50k vehicles). By integrating generative and contrastive self-supervised learning with context-aware sampling, NOVA significantly enhances IoV security, reliability, and adaptability.	10.1109/TNSM.2026.3696324
Jiayi Liu, Jinshuo Wang, Yizhi Huang, Chen Wang	LLM Deployment Strategies on Mobile Edge Servers for Dynamic Uncertain User Requests	2026	Early Access	Modeling Large language models Internet of Things	Leveraging on the task planning and solving capability of pretrained Large Language Models (LLMs), deploying LLM agents on Mobile Edge Computing (MEC) edge servers brings significant benefits for an Internet of Things (IoT) network for providing enhanced AI intelligence with acceptable delay. In this work, we consider the edge LLMs deployment strategy in an end-edge-cloud LLM agents system for the IoT services, which jointly determines the locations and number of LLM initializations and user requests offloading strategy in a dynamic network environment with stochastic user requests. We formulate this joint LLM Deployment and inference Tasks Offloading (LLMDTO) problem. Typically, we design an LLM service performance evaluation mechanism by measuring its processing delay with stochastic user requests arrivals by Stochastic Network Calculus (SNC). Due to the complexity of the LLMDTO problem, we decompose this joint optimization problem into two subproblems and propose an algorithm based on Multi Agent Deep Reinforcement Learning (MADRL) scheme. To accelerate the training process of the DRL, a reward model is designed by applying the Kolmogorov Arnold Networks (KAN) to return a fast reward estimation. Finally, we validate the proposed algorithm through extensive simulations and results show the effectiveness of the proposition on lower deployment cost and delay in a dynamic network environment.	10.1109/TNSM.2026.3708677