Last updated: 2025-07-18 03:01 UTC
All documents
Number of pages: 143
Author(s) | Title | Year | Publication | Keywords | ||
---|---|---|---|---|---|---|
Genxin Chen, Jin Qi, Jialin Hua, Ying Sun, Zhenjiang Dong, Yanfei Sun | AFAS: Arbitrary-Freedom Adaptive Scheduling for Multiworkflow Cloud Computing via Deep Reinforcement Learning | 2025 | Early Access | Processor scheduling Cloud computing Computational modeling Dynamic scheduling Optimization Deep reinforcement learning Adaptive scheduling Adaptation models Real-time systems Costs Cloud computing Deep reinforcement learning Multiple workflows Arbitrary freedom Adaptive scheduling | The in-depth development of artificial intelligence models has supported the high-quality allocation of cloud computing resources. The optimization of workflow scheduling issues in cloud computing has become increasingly critical due to the complexity of computing tasks, constraints on computing resources, and the growing demand for high-quality service. To address the increasingly complex workflow scheduling problems in cloud computing, this paper presents an arbitrary-freedom adaptive scheduling method for cloud computing with multiple workflows based on deep reinforcement learning (termed AFAS), with the workflow makespan and response time as the optimization objectives. First, we define the concept of degrees of freedom in the scheduling context to establish the feature space and foundational decision patterns relevant to multiworkflow scheduling. Second, an adaptive real-time scheduling strategy generation (ARS) algorithm is proposed for multiworkflow scheduling tasks. Third, a composite reward mechanism with an advanced-time-window real-time-reward (ATR) algorithm is designed for intelligent model optimization. Finally, the generation algorithm and intelligent model are fused to perform arbitrary-freedom multiworkflow adaptive scheduling. The experiments show that ATR can significantly increase the frequency of reward generation, AFAS can achieve at least 6.6% better performance than existing methods can achieve, and the incorporation of intelligent models improves the performance of ARS by 2.7%. | 10.1109/TNSM.2025.3566771 |
Dana Haj Hussein, Mohamed Ibnkahla | Towards Intelligent Intent-based Network Slicing for IoT Systems: Enabling Technologies, Challenges, and Vision | 2025 | Early Access | Internet of Things Surveys Resource management Translation Network slicing Ecosystems Data mining Autonomous networks Training Standardization Internet of Things Intent-based Networking Network Slicing Artificial Intelligent Machine Learning Resource Management Data Management | The rapid integration of intelligence and automation into future Internet of Things (IoT) systems, empowered by Intent-based Networking (IBN) and Network Slicing (NS) technologies, is transforming the way novel services are envisioned and delivered. The automation capabilities of IBN depend significantly on key facilitators, including data management and resource management. A robust data management methodology is essential for leveraging large-scale data, encompassing service-specific and network-specific data, enabling IBN systems to extract insights and facilitate real-time decision-making. Another critical enabler involves deploying intent-based mechanisms within an NS system that translate and ensure user intents by mapping them to precise Management and Orchestration (MO) commands. Nevertheless, data management in IoT systems faces significant security and operational challenges due to the diverse range of services and technologies involved. Furthermore, intent-based resource management demands intelligent proactive, and adaptive MO mechanisms that can fulfill a wide range of intent requirements. Existing surveys within the field have focused on technology-specific advancements, often overlooking these challenges. In response, this paper defines Intelligent Intent-Based Network Slicing (I-IBNS) systems exemplifying the integration of intelligent IBN and NS for the MO of IoT systems. Furthermore, the paper surveys I-IBNS systems, focusing on two critical domains: resource management and data management. The resource management segment examines recent developments in IBN mechanisms within an NS system. Meanwhile, the second segment explores data management complexities within IoT networks. Moreover, the paper envisions the roles of intent, NS, and the IoT ecosystem, thereby laying the foundation for future research directions. | 10.1109/TNSM.2025.3570052 |
Yi-Han Xu, Rong-Xing Ding, Xiao-Ren Xu, Ding Zhou, Wen Zhou | A DDPG Hybrid with Graph Scheme for Resource Management in Digital Twin-assisted Biomedical Cyber-Physical Systems | 2025 | Early Access | Resource management Reliability Biomedical monitoring Wireless communication Optimization Body area networks Real-time systems Monitoring Energy efficiency Vehicle dynamics digital twins biomedical cyber-physical system WBANs resource management Markov decision process | In this paper, we focus on the resource optimization issue in Wireless Body Area Networks (WBANs) section of a novel Biomedical Cyber Physical System (BM-CPS) in which the Digital Twin (DT) technique is adopted concurrently. We propose a scenario where multiple physiological sensor nodes continuously monitor the electrophysiology signals from patients and connect to a Cyber Managing Center (Cyb-MC) in a wireless way to transmit the physiological indices to the paramedic reliably. Specifically, to optimize the energy efficiency of physiological sensors while ensuring the reliability of emergency-critical electrophysiology signals transmissions and modeling the uncertain stochastic environments, the optimization issue is transformed into a Markov Decision Process (MDP) in terms of joint transmission mode, relay assignment, transmission power and time slot scheduling consideration. Subsequently, we propose a Random Graph-enabled Deep Deterministic Policy Gradient (RG-DDPG) scheme to tackle the challenge while releasing computing complexity. In addition, as an innovative paradigm for reshaping cyberspace applications, the DT of WBAN is created to capture the time-varying resource status, where virtualized and intelligent resource management can be performed uniformly. Finally, extensive simulation studies verified the advancement of the proposed scheme and demonstrated that the DT can mirror the topology, predict the behavior, and administrate the resources of the WBANs. | 10.1109/TNSM.2025.3570252 |
Cyril Shih-Huan Hsu, Jorge Martín-Pérez, Danny De Vleeschauwer, Luca Valcarenghi, Xi Li, Chrysa Papagianni | A Deep RL Approach on Task Placement and Scaling of Edge Resources for Cellular Vehicle-to-Network Service Provisioning | 2025 | Early Access | Delays Resource management Optimization Vehicle dynamics Urban areas Real-time systems Vehicle-to-everything Forecasting Deep reinforcement learning Transportation cellular vehicle to network task placement edge resource scaling deep reinforcement learning | Cellular Vehicle-to-Everything (C-V2X) is currently at the forefront of the digital transformation of our society. By enabling vehicles to communicate with each other and with the traffic environment using cellular networks, we redefine transportation, improving road safety and transportation services, increasing the efficiency of vehicular traffic flows, and reducing environmental impact. To effectively facilitate the provisioning of Cellular Vehicular-to-Network (C-V2N) services, we tackle the interdependent problems of service task placement and scaling of edge resources. Specifically, we formulate the joint problem and prove that it is not computationally tractable. To address its complexity we propose dhpg, a new Deep Reinforcement Learning (DRL) approach that operates in hybrid action spaces, enabling holistic decision-making and enhancing overall performance. We evaluated the performance of DHPG using simulations with a real-world C-V2N traffic dataset, comparing it to several state-of-the-art (SoA) solutions. DHPG outperforms these solutions, guaranteeing the 99th percentile of C-V2N service delay target, while simultaneously optimizing the utilization of computing resources. Finally, time complexity analysis is conducted to verify that the proposed approach can support real-time C-V2N services. | 10.1109/TNSM.2025.3570102 |
Xueqi Peng, Wenting Shen, Yang Yang, Xi Zhang | Secure Deduplication and Cloud Storage Auditing with Efficient Dynamic Ownership Management and Data Dynamics | 2025 | Early Access | Cloud computing Encryption Data integrity Vehicle dynamics Protocols Indexes Security Data privacy Servers Costs Cloud storage Integrity auditing Secure deduplication Ownership management Data dynamics | To verify the integrity of data stored in the cloud and improve storage efficiency, numerous cloud storage auditing schemes with deduplication have been proposed. In cloud storage, when users perform data dynamic operations, they should lose ownership of original data. However, existing schemes require re-encrypting the entire ciphertext when ownership changes and recalculating the authenticators for the blocks following the updated blocks when insertion or deletion operations are performed. These processes lead to high computation overhead. To address the above issues, we construct a secure deduplication and cloud storage auditing scheme with efficient dynamic ownership management and data dynamics. We adopt CAONT encryption method, where only a portion of the updated block is required to be re-encrypted during the ownership management phase, significantly reducing computation overhead. We also implement index switch sets to maintain the mapping between block indexes and cloud storage indexes of ciphertext blocks. By embedding cloud storage indexes within the authenticators, our scheme avoids the need to recalculate authenticators when users perform dynamic operations. Additionally, our scheme supports block-level deduplication, further improving efficiency. Through comprehensive security analysis and experiments, we validate the security and effectiveness of the proposed scheme. | 10.1109/TNSM.2025.3569833 |
Zhenxing Chen, Pan Gao, Teng-Fei Ding, Zhi-Wei Liu, Ming-Feng Ge | Practical Prescribed-Time Resource Allocation of NELAs with Event-Triggered Communication and Input Saturation | 2025 | Early Access | Resource management Event detection Costs Convergence Vehicle dynamics Mathematical models Training Perturbation methods Neurons Laplace equations Practical resource allocation networked Euler-Lagrange agents prescribed-time event-triggered communication input saturation | This paper investigates the practical resource allocation of networked Euler-Lagrange agents (NELAs) with event-triggered communication and input saturation. A novel prescribed-time resource allocation control (PTRAC) algorithm is proposed, which includes a resource allocation estimator and a prescribed-time NN-based local controller. The former is designed based on the time-based generator (TBG) and event-triggered mechanism to achieve the optimal resource allocation within the prescribed-time. Additionally, the prescribed-time NN-based local controller is designed using the approximation ability of RBF neural network to force the states of NELAs to track the optimal values within the prescribed-time. The most significant feature of the PTRAC algorithm is that the complex problem can be analyzed independently in chunks and converges within the prescribed-time, greatly reducing the number of triggers and communication costs. Subsequently, the validity is verified by simulation and several sufficient conditions are established via the Lyapunov stability argument. | 10.1109/TNSM.2025.3570091 |
Antonino Angi, Alessio Sacco, Guido Marchetto | LLNet: An Intent-Driven Approach to Instructing Softwarized Network Devices Using a Small Language Model | 2025 | Early Access | Translation Natural language processing Codes Accuracy Training Programming Pipelines Network topology Energy consumption Data mining user intents LLM SLM network programmability intent-based networking SDN | Traditional network management requires manual coding and expertise, making it challenging for non-specialists and experts to handle increasing devices and applications. In response, Intent-Based Networking (IBN) has been proposed to simplify network operations by allowing users to express in natural language the program objective (or intent), which is then translated into device-specific configurations. The emergence of Large Language Models (LLMs) has boosted the capabilities to interpret human intents, with recent IBN solutions embracing LLMs for a more accurate translation. However, while these solutions excel at intent comprehension, they lack a complete pipeline that can receive user intents and deploy network programs across devices programmed in multiple languages. In this paper, we present LLNet, our IBN solution that, within the context of Software-Defined Networking (SDN), can translate seamlessly intent-to-program. First, leveraging LLMs, we convert network intents into an intermediate representation by extracting key information; then, using this output, the system can tailor the network code for any topology using the specific language calls. At the same time, we address the challenge of a more sustainable IBN approach to reduce its energy consumption, and we experience how even a Small Language Model (SLM) can efficiently help LLNet for input translation. Results across multiple use cases demonstrated how our solution can guarantee adequate translation accuracy while reducing operator expenses compared to other LLM-based approaches. | 10.1109/TNSM.2025.3570017 |
Karcius D. R. Assis, Raul C. Almeida, Hojjat Baghban, Alex F. Santos, Raouf Boutaba | A Two-stage Reconfiguration in Network Function Virtualization: Toward Service Function Chain Optimization | 2025 | Early Access | Resource management Optimization Service function chaining Substrates Network function virtualization Servers Scalability Real-time systems Virtual machines Routing Service Function Chain Reconfiguration Optimization Network Function Virtualization VNF Migration | Network Function Virtualization (NFV), as a promising paradigm, speeds up the service deployment by separating network functions from proprietary devices and deploying them on common servers in the form of software. Any service in NFV-enabled networks is achieved as a Service Function Chain (SFC) which consists of a series of ordered Virtual Network Functions (VNFs). However, migration of VNFs for more flexible services within a dynamic NFV-enabled network is a key challenge to be addressed. Current VNF migration studies mainly focus on single VNF migration decisions without considering the sharing and concurrent migration of VNF instances. In this paper, we assume that each deployed VNF is used by multiple SFCs and deal with the optimal placement for the contemporaneous migration of VNFs based on the actual network situation. We formalize the VNF migration and SFC reconfiguration problem as a mathematical model, which aims to minimize the VNF migration between nodes or the total number of core changes per node. The approach is a two-stage MILP based on optimal order to solve the reconfiguration. Extensive evaluation shows that the proposed approach can reduce the change in terms of location or number of cores per node in a 6-node and 14-node networks while ensuring network latency compared with the model without reconfiguration. | 10.1109/TNSM.2025.3567906 |
Jinshui Wang, Yao Xin, Chongwu Dong, Lingfeng Qu, Yiming Ding | ERPC: Efficient Rule Partitioning Through Community Detection for Packet Classification | 2025 | Early Access | Decision trees Classification algorithms Partitioning algorithms Manuals Tuning Optimization Throughput IP networks Memory management Vectors Packet Classification rule partitioning decision tree community detection graph coloring | Packet classification is crucial for network security, traffic management, and quality of service by enabling efficient identification and handling of data packets. Decision tree-based rule partitioning has emerged as a prominent method in recent research. A significant challenge for decision tree algorithms is rule replication, which occurs when rules span multiple subspaces, leading to substantial memory consumption increases. Rule partitioning can effectively mitigate or eliminate this replication by separating overlapping rules. However, existing partitioning techniques heavily rely on manual parameter tuning across a wide range of possible values, making optimal solution discovery challenging. Furthermore, due to the lack of global optimization, these approaches face a critical trade-off: either the number of subsets becomes uncontrollable, resulting in diminished query speed, or rule replication becomes severe, causing substantial memory overhead. To bridge these gaps and achieve high-performance adaptive partitioning, we propose ERPC, a novel algorithm with the following key features: First, ERPC leverages graph theory to model rule sets, enabling global optimization that balances intra-group rule replication against the total number of groups. Second, ERPC advances rule set partitioning by modifying traditional community detection algorithms, strategically shifting the optimization objective from positive to negative modularity. Third, ERPC allows the rule set itself to determine the optimal number of groups, thus eliminating the need for manual parameter tuning. Experimental results demonstrate the efficacy of ERPC when applied to CutSplit, a state-of-the-art multi-tree method. It preserves 88% of CutSplit’s average classification throughput while reducing tree-building time by 89% and memory consumption by 77%. Furthermore, ERPC exhibits strong scalability, being adaptable to mainstream decision tree methods. | 10.1109/TNSM.2025.3567705 |
Lingling Wang, Zhengyin Zhang, Mei Huang, Keke Gai, Jingjing Wang, Yulong Shen | RoPA: Robust Privacy-Preserving Forward Aggregation for Split Vertical Federated Learning | 2025 | Early Access | Vectors Protocols Privacy Training Data models Cryptography Resists Arithmetic Protection Federated learning Vertical federated learning Robust Privacy-preserving Integrity SNIP | Split Vertical Federated Learning (Split VFL) is an increasingly popular framework for collaborative machine learning on vertically partitioned data. However, it is vulnerable to various attacks, resulting in privacy leakage and robust aggregation issues. Recent works have explored the privacy protection of raw data samples and labels, neglecting malicious attacks launched by dishonest passive parties. Since they may deviate from the protocol and launch embedding poisoning attacks and free-riding attacks, it will inevitably result in model performance loss. To address this issue, we propose a Robust Privacy-preserving forward Aggregation (RoPA) protocol, which can resist embedding poisoning attacks and free-riding attacks and protect the privacy of embedding vectors. Specifically, we first present a modified Secret-shared Non-Interactive Proofs (SNIP) algorithm to guarantee the integrity verification of embedding vectors. To prevent free-riding attacks, we also give a validity verification protocol using matrix commitment. In particular, we utilize probability checking and batch verification to improve the verification efficiency of the protocol. Moreover, we adopt arithmetic secret sharing to protect data privacy. Finally, we conduct rigorous theoretical analysis to prove the security of RoPA and evaluate the performance of RoPA. The experimental results show that the proof verification overhead of RoPA is approximately 8× lower than the original SNIP, and the model accuracy is improved by ranging from 3% to 15% under the above two malicious attacks. | 10.1109/TNSM.2025.3569228 |
Ruopeng Geng, Jianyuan Lu, Chongrong Fang, Shaokai Zhang, Jiangu Zhao, Zhigang Zong, Biao Lyu, Shunmin Zhu, Peng Cheng, Jiming Chen | Enabling Stateful TCP Performance Profiling with Key Event Capturing | 2025 | Early Access | Kernel Production Probes Filtering Training Servers Packet loss Linux Electronic mail Data communication Network Measurement TCP Performance Profiling TCP Stack Probes | TCP ensures reliable transmission through its stateful implementation and remains crucial today. TCP performance profiling is essential for tasks like diagnosing network performance problems, optimizing transmission performance, and developing new TCP variants, etc. Existing profiling methods lack enough attention to TCP state transition to provide detailed insights on TCP performance. Thus, we build TcpSight, a tool focusing on TCP state transition throughout connection lifetimes. TcpSight conducts stateful analysis by capturing key events using an efficient per-connection lock-free data management mechanism. Besides, TcpSight enhances profiling by integrating application layer information collected from the TCP stack. With the profiling results, users can identify the culprit of TCP performance degradation, and evaluate the performance of TCP algorithms. We design optional modules and filtering mechanisms to reduce TcpSights overhead. Our evaluation presents that TcpSight incurs an additional CPU consumption of about 16.6% (without filtering) and 10.6% (with filtering) when the servers load is 55.7%, and generates storage consumption about 1.88 KB per connection on average. We also give application cases of TcpSight and the deployment experiences in Alibaba Cloud. TcpSight helps in revealing meaningful findings and insights into exploiting TCP in the production deployment. | 10.1109/TNSM.2025.3564336 |
Wenjing Gao, Jia Yu | Enabling Privacy-Preserving Top-k Hamming Distance Query on the Cloud | 2025 | Early Access | Hamming distances Protocols Cloud computing Servers Encryption Games Computational modeling Social networking (online) Privacy Social groups Cloud computing cloud security privacy preserving Hamming distance top-k query | The top-k Hamming distance query is to find the k optimal objects with the smallest Hamming distance to the query data. It has a wide range of applications in many domains such as social networks, image retrieval and biological recognition. The existing privacy-preserving protocols do not support the top-k Hamming distance query in practice. To address this issue, we consider letting the user securely query the top-k Hamming distance on the cloud in a secure outsourcing manner. We propose two protocols to realize the privacy-preserving top-k Hamming distance query on the cloud. In the first protocol, two cloud servers are introduced to cooperatively complete the privacy-preserving top-k Hamming distance query. To preserve data privacy, the Paillier encryption and randomization techniques are leveraged to blind the user data, and the ciphertext data is stored on the first cloud server. The second cloud server calculates the Hamming distance on the ciphertexts. After that, the encrypted query results are returned to the query user for recovering the top-k query results. In the second protocol, we adopt the data aggregation strategy to further enhance the efficiency. By packaging data, the computation overhead of each participant is reduced and the communication overhead of the protocol is decreased, remarkably. Security analysis demonstrates that the data privacy is guaranteed in the proposed protocols. Experimental results evaluate the performance of the proposed protocols and confirm the superiority of the second protocol. | 10.1109/TNSM.2025.3565943 |
Chao Zha, Zhiyu Wang, Yifei Fan, Bing Bai, Yinjie Zhang, Sainan Shi, Ruyun Zhang | DM-IDS -A Network Intrusion Detection Method Based on Dual-Modal Fusion | 2025 | Early Access | Feature extraction Payloads Network intrusion detection Telecommunication traffic Vectors Graph neural networks Radio frequency Data mining Training Technological innovation Intrusion detection flow modal payload modal bilinear fusion semantic | The machine learning-based approach to network intrusion detection presents a groundbreaking research paradigm, positioned to replace traditional rule-based and signature-based methods. However, prior research methodologies have predominantly focused on flow-based approaches, which may not be effective in detecting all types of attacks at a granular level. In this study, we introduce DM-IDS, an attention-convolution architecture model for bimodal network intrusion detection in both flow and payload modalities, using bilinear fusion. Notably, we present a novel method for constructing binary-form feature vectors under the payload modality, with the goal of extracting additional security semantic features. To facilitate this, we independently develop a feature generation tool named Beeman. Finally, we conduct a series of comparative and ablation experiments on two publicly available datasets, CICIDS-2017 and CICIoT-2023, achieving state-of-the-art model performance. | 10.1109/TNSM.2025.3565614 |
Vaishnavi Kasuluru, Luis Blanco, Engin Zeydan | Enhancing Open RAN Operations: The Role of Probabilistic Forecasting in Network Analysis | 2025 | Early Access | Probabilistic logic Open RAN Forecasting Resource management Predictive models Heuristic algorithms Computer architecture Adaptation models Dynamic scheduling Uncertainty Open RAN 6G Probabilistic Forecasting Network Analytics AI | Resource provisioning plays a crucial role in effective resource management. As we move into the 6G era, technologies such as Open Radio Access Network (O-RAN) offer the opportunity to develop intelligent and interoperable cutting-edge solutions for qualitative management of the latest communication system. Previous works have mostly used single-point forecasts like Long-Short Term Memory (LSTM) for predicting resource requirements, which presents decision-makers with the problem of making informed decisions about resource allocation. On the other hand, probability-based forecasting techniques such as DeepAR, Transformer and Simple-Feed-Forward (SFF) offer new dimensions to the predictions by quantifying their uncertainties. This work shows the comprehensive comparison of single-point and probabilistic estimators and evaluates their effectiveness in predicting the actual number of Physical Resource Blocks (PRBs) needed in the context of O-RAN, especially for multi-tenant use cases. The results show the superiority of the probabilistic model in terms of various evaluation metrics. DeepAR achieves the highest accuracy, outperforming single-point and other probabilistic estimators. Based on these findings, a novel approach named Dynamic Percentile Adjustment Approach (DYNp) algorithm is proposed, which utilizes probabilistic forecasting for adaptive resource allocation. After extensive analysis, the numerical results show that the DYNp algorithm for DeepAR predictions reduces the Service Level Agreement (SLA) violation to 8% and the over-provisioning to 0.509 by dynamic percentile adaption. DYNp approach ensures that resources are allocated by efficiently handling over-and under-provisioning, making it suitable for real-time scenarios in O-RAN environments. | 10.1109/TNSM.2025.3565268 |
Pavlos S. Bouzinis, Panagiotis Radoglou-Grammatikis, Ioannis Makris, Thomas Lagkas, Vasileios Argyriou, Georgios Th. Papadopoulos, Panagiotis Sarigiannidis, George K. Karagiannidis | StatAvg: Mitigating Data Heterogeneity in Federated Learning for Intrusion Detection Systems | 2025 | Early Access | Servers Training Intrusion detection Data models Europe Feature extraction Privacy Data augmentation Convolutional neural networks Batch normalization cybersecurity intrusion detection systems federated learning data heterogeneity statistical averaging | Federated learning (FL) enables devices to collaboratively build a shared machine learning (ML) or deep learning (DL) model without exposing raw data. Its privacy-preserving nature has made it popular for intrusion detection systems (IDS) in the field of cybersecurity. However, data heterogeneity across participants poses challenges for FL-based IDS. This paper proposes statistical averaging (StatAvg) method to alleviate non-independently and identically (non-iid) distributed features across local clients’ data in FL. In particular, StatAvg allows the FL clients to share their individual local data statistics with the server. These statistics include the mean and variance of each client’s feature vector. The server then aggregates this information to produce global statistics, which are shared with the clients and used for universal data normalization, i.e., common scaling of the input features by all clients. It is worth mentioning that StatAvg can seamlessly integrate with any FL aggregation strategy, as it occurs before the actual FL training process. The proposed method is evaluated against well-known baseline approaches that rely on batch and layer normalization, such as FedBN, and address the non-iid features issue in FL. Experiments were conducted using the TON-IoT and CIC-IoT-2023 datasets, which are relevant to the design of host and network IDS, respectively. The experimental results demonstrate the efficiency of StatAvg in mitigating non-iid feature distributions across the FL clients compared to the baseline methods, offering a gain in IDS accuracy ranging from 4% to 17%. | 10.1109/TNSM.2025.3564387 |
Ying-Dar Lin, Yin-Tao Ling, Yuan-Cheng Lai, Didik Sudyana | Reinforcement Learning for AI as a Service: CPU-GPU Task Scheduling for Preprocessing, Training, and Inference Tasks | 2025 | Early Access | Artificial intelligence Graphics processing units Training Computer architecture Optimal scheduling Scheduling Real-time systems Complexity theory Resource management Inference algorithms AI as a Service CPU GPU Task Scheduling Reinforcement Learning Deep Q-Learning | The rise of AI solutions has driven the emergence of AI as a Service (AIaaS), offering cost-effective and scalable solutions by outsourcing AI functionalities to specialized providers. Within AIaaS, three key components are essential: segmenting AI services into preprocessing, training, and inference tasks; utilizing GPU-CPU heterogeneous systems where GPUs handle parallel processing and CPUs manage sequential tasks; and minimizing latency in a distributed architecture consisting of cloud, edge, and fog computing. Efficient task scheduling is crucial to optimize performance across these components. In order to enhance task scheduling in AIaaS, we propose a user-experience-and-performance-balanced reinforcement learning (UXP-RL) algorithm. The UXP-RL algorithm considers 11 factors, including queuing task information. It then estimates resource release times and observes previous action outcomes, to select the optimal AI task for execution on either a GPU or CPU. This method effectively reduces the average turnaround time, particularly for rapid inference tasks. Our experimental findings show that the proposed RL-based scheduling algorithm reduces average turnaround time by 27.66% to 57.81% compared to the heuristic approaches such as SJF and FCFS. In a distributed architecture, utilizing distributed RL schedulers reduces the average turnaround time by 89.07% compared to a centralized scheduler. | 10.1109/TNSM.2025.3564480 |
Jin-Xian Liu, Jenq-Shiou Leu | ETCN-NNC-LB: Ensemble TCNs With L-BFGS-B Optimized No Negative Constraint-Based Forecasting for Network Traffic | 2025 | Early Access | Telecommunication traffic Forecasting Predictive models Data models Convolutional neural networks Long short term memory Ensemble learning Complexity theory Accuracy Overfitting Deep learning ensemble learning network traffic prediction temporal convolutional network (TCN) time series forecasting | With the increasing demand for internet access and the advent of technologies such as 6G and IoT, efficient and dynamic network resource management has become crucial. Accurate network traffic prediction plays a pivotal role in this context. However, existing prediction methods often struggle with challenges such as complexity-accuracy trade-offs, limited data availability, and diverse traffic patterns, especially in coarsegrained forecasting. To address these issues, this article proposes ETCN-NNC-LB, which is a novel ensemble learning method for network traffic forecasting. ETCN-NNC-LB combines Temporal Convolutional Networks (TCNs) with No Negative Constraint Theory (NNCT) weight integration in ensemble learning and is optimized using the Limited-memory Broyden-Fletcher-Goldfarb- Shanno with Box constraints (L-BFGS-B) algorithm. This method balances model complexity and accuracy, mitigates overfitting risks, and flexibly aggregates predictions. The ETCN-NNC-LB model also incorporates a pattern-handling method to forecast traffic behaviors robustly. Experiments on a real-world dataset demonstrate that ETCN-NNC-LB significantly outperforms stateof-the-art methods, achieving an approximately 22% reduction in the Root Mean Square Error (RMSE). The proposed method provides accurate and efficient network traffic prediction in dynamic, data-constrained environments. | 10.1109/TNSM.2025.3563978 |
Wenxian Li, Pingang Cheng, Yong Feng, Nianbo Liu, Ming Liu, Yingna Li | A Blockchain-Assisted Hierarchical Data Aggregation Framework for IIoT With Computing First Networks | 2025 | Early Access | Industrial Internet of Things Blockchains Data privacy Data aggregation Cloud computing Collaboration Servers Edge computing Resource management Protection Industrial internet of things(IIoT) computing first networks(CFN) secure data aggregation blockchain | With an increasing number of sensor devices connected to industrial systems, the efficient and reliable aggregation of sensor data has become a key topic in Industrial Internet of Things (IIoT). Computing First Networks (CFN) are emerging as a promising technology for aggregating vast quantities of IIoT data. However, existing CFN data collection frameworks are usually centralized, which overly rely on third-party trusted authorities and fail to fully schedule and utilize limited computing resources. More critically, that is prone to trust and security issues. In this paper, considering the heterogeneity and data security in complex industrial scenarios, we propose a blockchain-based and multi-edge CFN collaborative IIoT data hierarchical collection framework (ME-CIDC) to collect massive IIoT data securely and efficiently. In ME-CIDC, a blockchain-driven resource allocation algorithm is proposed for inter-domain CFN, which achieves distributed and efficient task scheduling and data collection by constructing multiple blockchains. A self-incentive mechanism is designed to encourage inter-domain nodes to contribute resources and support the operation of the inter-domain CFN. We also propose an efficient double-layered data aggregation algorithm, which distributes computational tasks across two layers to ensure the efficient collection and aggregation of IIoT data. Extensive simulation and numerical results demonstrate the effectiveness of our proposed scheme. | 10.1109/TNSM.2025.3563237 |
Jing Zhu, Dan Wang, Jiao Xing, Shuxin Qin, Gaofeng Tao, Pingping Chen, Zuqing Zhu | Bilevel Optimization for Provisioning Heterogeneous Traffic in Deterministic Networks | 2025 | Early Access | Optimization Bandwidth Routing Quality of service Bars Channel allocation Job shop scheduling IP networks Approximation algorithms Switches Deterministic networking Normal and deterministic traffic Bandwidth allocation routing and scheduling Bilevel optimization | Due to the capabilities of providing extremely low packet loss and bounded end-to-end latency, deterministic networking (DetNet) has been considered as a promising technology for emerging time-sensitive applications (e.g., industrial control and smart grids) in IP networks. To provide deterministic services, the operator needs to address the routing and scheduling problem. In this work, we study the problem from a novel prospective, i.e., the problem should be optimized not only for deterministic traffic, but also for normal traffic to coexist with the former. Specifically, we redefine the problem as bandwidth allocation, routing and scheduling (BaRS), and model this problem as a bilevel optimization which consists of an upper-level optimization and a lower-level optimization. The upper-level optimization allocates link bandwidth between deterministic and normal traffic to maximize the available bandwidth for normal traffic on the premise of accepting a certain portion of deterministic bandwidth; the lower-level optimization determines specific routing and scheduling solutions for deterministic traffic to maximize the number of accepted deterministic flows. We first formulate the bilevel optimization as a bilevel mixed integer linear programming (BMILP). Then, we propose an exact algorithm based on cutting planes to solve it exactly, and propose an approximation algorithm based on two-level relaxations and randomized rounding to solve it effectively and time-efficiently. Extensive simulations are conducted and the results verify the effectiveness of our proposals in balancing the tradeoff between the available bandwidth for normal traffic and the number of accepted deterministic flows. | 10.1109/TNSM.2025.3570284 |
Cristian Zilli, Alessio Sacco, Flavio Esposito, Guido Marchetto | ClearNET: Enhancing Transparency in Opaque Network Models using eXplainable AI (XAI) for Efficient Traffic Engineering | 2025 | Early Access | Computational modeling Data models Feature extraction Explainable AI Telemetry Predictive models Training Telecommunication traffic Routing Deep learning XAI traffic engineering machine learning | AI/ML has enhanced computer networking, aiding administrators in decision-making and automating tasks for optimized performance. Despite such advances in network automation, there remains limited trust in these uninterpretable models due to their inherent complexity. To this aim, eXplainable AI (XAI) has emerged as a critical area to demystify (deep) neural network models and to provide more transparent decision-making processes. While other fields have embraced XAI more prominently, the use of these techniques in computer network management remains largely unexplored. In this paper, we shed some light by presenting, an XAI-based approach designed to clarify the opaque nature of data-driven traffic engineering solutions in general, and efficient network telemetry, in particular. It does so by examining the intrinsic behavior of the adopted models, thereby reducing the volume of data needed for effective learning. Our extensive evaluation revealed how our approach not only reduces training time and overhead in network telemetry models but also maintains or improves model accuracy, leading, in turn, to more efficient and clear ML models for network management. | 10.1109/TNSM.2025.3567654 |