Pub Date : 2026-01-19DOI: 10.1109/TNSE.2026.3655675
Meilin Xu;Min Jia;Yuyan Ren;Qing Guo;Tomaso de Cola
With the widespread deployment of low earth orbit (LEO) satellite networks, their high dynamism and large-scale introduce new challenges for the management and control of network communication resources and service orchestration. To tackle these challenges, this paper leverages software defined networking (SDN) and Network Function Virtualization (NFV) to the joint optimization of virtualized network function (VNF) deployment and request scheduling, referred to as the Joint VNF Deployment and Scheduling problem for Mobile Satellite Networks (JVDS-MSN). We formulate the JVDS-MSN problem as an Integer Linear Programming model with cross-timeslot service continuity constraints, aiming to minimize the end-to-end communication resource consumption. Given the NP-hard nature of the problem, we first propose an exact optimization method that integrates Dantzig-Wolfe decomposition with branch-and-bound techniques (DW-BP) to obtain optimal solutions. Although the proposed DW-BP algorithm yields high-quality solutions, its computational cost limits its applicability to large-scale scenarios. To address this, we propose a hierarchical reinforcement learning algorithm based on Twin Delayed Deep Deterministic Policy Gradient (HRL-TD3). This algorithm decomposes the VNF deployment and request scheduling tasks into high-level and low-level sub-tasks, thereby enabling more efficient optimization of bandwidth resources. Simulation results show that the proposed DW-BP algorithm efficiently computes optimal solutions, serving as a strong performance baseline. In large-scale and heterogeneous satellite network scenarios, the HRL-TD3 algorithm achieves near-optimal performance with significantly reduced computational overhead. Overall, the proposed method offers a promising solution for scalable and efficient service orchestration in mobile satellite networks.
{"title":"Joint Optimization of VNF Deployment and Request Scheduling in Mobile Satellite Networks","authors":"Meilin Xu;Min Jia;Yuyan Ren;Qing Guo;Tomaso de Cola","doi":"10.1109/TNSE.2026.3655675","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3655675","url":null,"abstract":"With the widespread deployment of low earth orbit (LEO) satellite networks, their high dynamism and large-scale introduce new challenges for the management and control of network communication resources and service orchestration. To tackle these challenges, this paper leverages software defined networking (SDN) and Network Function Virtualization (NFV) to the joint optimization of virtualized network function (VNF) deployment and request scheduling, referred to as the Joint VNF Deployment and Scheduling problem for Mobile Satellite Networks (JVDS-MSN). We formulate the JVDS-MSN problem as an Integer Linear Programming model with cross-timeslot service continuity constraints, aiming to minimize the end-to-end communication resource consumption. Given the NP-hard nature of the problem, we first propose an exact optimization method that integrates Dantzig-Wolfe decomposition with branch-and-bound techniques (DW-BP) to obtain optimal solutions. Although the proposed DW-BP algorithm yields high-quality solutions, its computational cost limits its applicability to large-scale scenarios. To address this, we propose a hierarchical reinforcement learning algorithm based on Twin Delayed Deep Deterministic Policy Gradient (HRL-TD3). This algorithm decomposes the VNF deployment and request scheduling tasks into high-level and low-level sub-tasks, thereby enabling more efficient optimization of bandwidth resources. Simulation results show that the proposed DW-BP algorithm efficiently computes optimal solutions, serving as a strong performance baseline. In large-scale and heterogeneous satellite network scenarios, the HRL-TD3 algorithm achieves near-optimal performance with significantly reduced computational overhead. Overall, the proposed method offers a promising solution for scalable and efficient service orchestration in mobile satellite networks.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"6106-6121"},"PeriodicalIF":7.9,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082123","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-16DOI: 10.1109/TNSE.2026.3654756
Yuxing Zhang;Lingling Wang;Meng Li;Keke Gai;Jingjing Wang
Byzantine-robust Federated Learning (FL) enables service providers to learn an accurate global model, even when some participants may be malicious. Existing Byzantine-robust FL approaches primarily rely on the service provider conducting statistical analysis on clients’ model updates, filtering out anomalous ones before aggregation to refine the global model. However, these defenses struggle to distinguish benign outliers from anomalous model updates under Byzantine attacks and heterogeneous settings, thereby harming model generalization ability. To address this issue, we propose a Byzantine-robust aggregation scheme based on hybrid anomaly detection (HadAGG) in heterogeneous FL. Specifically, we introduce a hybrid filtering strategy combining cosine similarity and Shapley values to distinguish between benign, malicious, and anomalous but benign model updates. To effectively identify benign outliers, we propose a Shapley value-based approach by constructing a multi-objective utility function that integrates the loss function and model accuracy to compute the Federated Shapley value, which measures client contributions. To achieve Byzantine-robust aggregation, we correct malicious model updates via gradient projection instead of directly discarding them, and employ a weighted aggregation to ensure that all model updates have a positive effect on model performance. Finally, we perform a theoretical analysis and a comprehensive evaluation for our scheme. Experimental results show that HadAGG outperforms existing state-of-the-art (SOTA) Byzantine-robust aggregation methods under different attack scenarios.
{"title":"“Malicious or Benign?”: Enhancing the Contribution of Model Updates in Byzantine-Robust Heterogeneous Federated Learning","authors":"Yuxing Zhang;Lingling Wang;Meng Li;Keke Gai;Jingjing Wang","doi":"10.1109/TNSE.2026.3654756","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3654756","url":null,"abstract":"Byzantine-robust Federated Learning (FL) enables service providers to learn an accurate global model, even when some participants may be malicious. Existing Byzantine-robust FL approaches primarily rely on the service provider conducting statistical analysis on clients’ model updates, filtering out anomalous ones before aggregation to refine the global model. However, these defenses struggle to distinguish benign outliers from anomalous model updates under Byzantine attacks and heterogeneous settings, thereby harming model generalization ability. To address this issue, we propose a Byzantine-robust aggregation scheme based on hybrid anomaly detection (HadAGG) in heterogeneous FL. Specifically, we introduce a hybrid filtering strategy combining cosine similarity and Shapley values to distinguish between benign, malicious, and anomalous but benign model updates. To effectively identify benign outliers, we propose a Shapley value-based approach by constructing a multi-objective utility function that integrates the loss function and model accuracy to compute the Federated Shapley value, which measures client contributions. To achieve Byzantine-robust aggregation, we correct malicious model updates via gradient projection instead of directly discarding them, and employ a weighted aggregation to ensure that all model updates have a positive effect on model performance. Finally, we perform a theoretical analysis and a comprehensive evaluation for our scheme. Experimental results show that HadAGG outperforms existing state-of-the-art (SOTA) Byzantine-robust aggregation methods under different attack scenarios.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"6027-6040"},"PeriodicalIF":7.9,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082007","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Cyber attacks pose serious threats to computer systems. Automatically detecting anomalous patterns in system logs is critical for identifying and mitigating security risks. However, as log data grows increasingly complex and labeled logs remain scarce, existing detection methods face significant challenges. To address these issues, we introduce the pre-training and fine-tuning paradigm for log analysis and propose a hybrid pipeline tailored for accurate and low-cost log anomaly detection. Specifically, we employ a masked log reconstruction strategy to pre-train a Transformer encoder–based foundation model by leveraging the sequential dependencies in unlabeled logs. The model is then fine-tuned on an event prediction task to derive the anomaly detector. To reduce computational and storage overhead, we further design a knowledge distillation method tailored for compressing log anomaly detectors. Beyond fitting the detector's outputs, our method also exploits its internal representations to transfer richer knowledge. Experiments on the HDFS, BGL, and Thunderbird public datasets demonstrate that our framework outperforms state-of-the-art baselines in multiple metrics. Empirical evaluation on a reconstructed HDFS dataset confirms that it can adapt to real-world scenarios where labeled data is scarce. Moreover, through our knowledge distillation approach, the lightweight detectors achieve outstanding performance with substantially lower overhead, while maintaining robustness in real-world scenarios.
{"title":"Log Anomaly Detection via Transformers Pre-Trained on Massive Unlabeled Data","authors":"Senming Yan;Lei Shi;Jing Ren;Wei Wang;Limin Sun;Wei Zhang","doi":"10.1109/TNSE.2026.3654089","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3654089","url":null,"abstract":"Cyber attacks pose serious threats to computer systems. Automatically detecting anomalous patterns in system logs is critical for identifying and mitigating security risks. However, as log data grows increasingly complex and labeled logs remain scarce, existing detection methods face significant challenges. To address these issues, we introduce the pre-training and fine-tuning paradigm for log analysis and propose a hybrid pipeline tailored for accurate and low-cost log anomaly detection. Specifically, we employ a masked log reconstruction strategy to pre-train a Transformer encoder–based foundation model by leveraging the sequential dependencies in unlabeled logs. The model is then fine-tuned on an event prediction task to derive the anomaly detector. To reduce computational and storage overhead, we further design a knowledge distillation method tailored for compressing log anomaly detectors. Beyond fitting the detector's outputs, our method also exploits its internal representations to transfer richer knowledge. Experiments on the HDFS, BGL, and Thunderbird public datasets demonstrate that our framework outperforms state-of-the-art baselines in multiple metrics. Empirical evaluation on a reconstructed HDFS dataset confirms that it can adapt to real-world scenarios where labeled data is scarce. Moreover, through our knowledge distillation approach, the lightweight detectors achieve outstanding performance with substantially lower overhead, while maintaining robustness in real-world scenarios.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5943-5960"},"PeriodicalIF":7.9,"publicationDate":"2026-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082078","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1109/TNSE.2026.3654163
Long Luo;Yanan Huang;Xixi Chen;Yongsheng Zhao;Hongfang Yu;Schahram Dustdar
In-network aggregation (INA) accelerates gradient aggregation in distributed machine learning (DML) by alleviating communication bottlenecks, but its effectiveness crucially depends on two location decisions: where to deploy INA functions and where to aggregate gradient flows. Most existing methods optimize INA placement and gradient flow routing independently, missing the advantages of joint optimization. This paper presents LLMINA, which leverages Large Language Models (LLMs) to automate the heuristic design for joint INA placement and gradient aggregation, aiming to minimize makespan (i.e., the total time required for all DML jobs to complete gradient aggregation). Directly using LLMs to generate end-to-end solutions is infeasible due to problem complexity and LLM limitations. Instead, LLMINA uses LLMs to generate heuristics for INA placement through an evolutionary process, and then applies an optimization-based heuristic for gradient routing that takes into account DML workload characteristics. Experiments across diverse network topologies and workloads show that LLMINA can significantly reduce makespan compared to state-of-the-art baselines. These results underscore that location matters for both INA deployment and aggregation, and highlight the potential of LLM-guided heuristic design for complex network resource optimization.
{"title":"Location Matters: LLM-Guided Joint Optimization of In-Network Aggregation Placement and Routing for DML Workloads","authors":"Long Luo;Yanan Huang;Xixi Chen;Yongsheng Zhao;Hongfang Yu;Schahram Dustdar","doi":"10.1109/TNSE.2026.3654163","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3654163","url":null,"abstract":"In-network aggregation (INA) accelerates gradient aggregation in distributed machine learning (DML) by alleviating communication bottlenecks, but its effectiveness crucially depends on two location decisions: where to deploy INA functions and where to aggregate gradient flows. Most existing methods optimize INA placement and gradient flow routing independently, missing the advantages of joint optimization. This paper presents LLMINA, which leverages Large Language Models (LLMs) to automate the heuristic design for joint INA placement and gradient aggregation, aiming to minimize makespan (i.e., the total time required for all DML jobs to complete gradient aggregation). Directly using LLMs to generate end-to-end solutions is infeasible due to problem complexity and LLM limitations. Instead, LLMINA uses LLMs to generate heuristics for INA placement through an evolutionary process, and then applies an optimization-based heuristic for gradient routing that takes into account DML workload characteristics. Experiments across diverse network topologies and workloads show that LLMINA can significantly reduce makespan compared to state-of-the-art baselines. These results underscore that location matters for both INA deployment and aggregation, and highlight the potential of LLM-guided heuristic design for complex network resource optimization.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5978-5991"},"PeriodicalIF":7.9,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081967","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1109/TNSE.2026.3654107
Runxiao Liu;Xiangli Le;Shuang Gu;Shuli Lv;Pengda Mao;Quan Quan
This paper presents a novel distributed network control framework for cooperative heavy-load transportation using multi-UAV systems, accounting for thrust limitations and heterogeneous cable characteristics. By constructing a virtual passive system comprising interconnected virtual nodes, springs, and dampers, the proposed method decouples internal coordination stability from external velocity tracking. A velocity tracking controller is devised to asymptotically steer the load’s velocity toward a desired trajectory, while preserving inter-agent cohesion through virtual interactions. Notably, the controller operates without explicit inter-UAV communication, relying solely on relative position measurements. Numerical simulations involving ten UAVs transporting a 14 kg load-exceeding 76% of their combined thrust capacity-along a figure-eight trajectory validate the proposed method. Field tests with six UAVs transporting a 6 kg load are conducted to validate the control framework’s performance in practical applications. The results confirm accurate velocity tracking, balanced cable tension distribution, and scalability to heterogeneous UAV team configurations.
{"title":"Distributed Network Control of Multi-UAV Systems for Cooperative Heavy-Load Transport Using a Virtual-Passivity Framework","authors":"Runxiao Liu;Xiangli Le;Shuang Gu;Shuli Lv;Pengda Mao;Quan Quan","doi":"10.1109/TNSE.2026.3654107","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3654107","url":null,"abstract":"This paper presents a novel distributed network control framework for cooperative heavy-load transportation using multi-UAV systems, accounting for thrust limitations and heterogeneous cable characteristics. By constructing a virtual passive system comprising interconnected virtual nodes, springs, and dampers, the proposed method decouples internal coordination stability from external velocity tracking. A velocity tracking controller is devised to asymptotically steer the load’s velocity toward a desired trajectory, while preserving inter-agent cohesion through virtual interactions. Notably, the controller operates without explicit inter-UAV communication, relying solely on relative position measurements. Numerical simulations involving ten UAVs transporting a 14 kg load-exceeding 76% of their combined thrust capacity-along a figure-eight trajectory validate the proposed method. Field tests with six UAVs transporting a 6 kg load are conducted to validate the control framework’s performance in practical applications. The results confirm accurate velocity tracking, balanced cable tension distribution, and scalability to heterogeneous UAV team configurations.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5905-5923"},"PeriodicalIF":7.9,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082081","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recently, reconfigurable intelligent surface (RIS) has been widely used to enhance the performance of millimeter wave (mmWave) communication systems, making beam alignment more challenging. To ensure efficient communication, this paper proposes a novel intelligent angle map-based beam alignment scheme for both general user equipments (UEs) and RIS-aided UEs simultaneously in a fast and effective way. Specifically, we construct a beam alignment architecture that utilizes only angular information. To obtain the angle information, the currently hottest seq2seq model – the Transformer – is introduced to offline learn the relationship between UE geographic location and the corresponding optimal beam direction. Based on the powerful machine learning model, the location-angle mapping function, i.e., the angle map, can be built. As long as the location information of UEs is available, the angle map can make the acquisition of beam alignment angles effortless. In the simulation, we utilize a ray-tracing-based dataset to verify the performance of the proposed scheme. It is demonstrated that the proposed scheme can achieve high-precision beam alignment and remarkable system performance without any beam scanning.
{"title":"Intelligent Angle Map-Based Beam Alignment for RIS-Aided mmWave Communication Networks","authors":"Hao Xia;Qing Xue;Yanping Liu;Binggui Zhou;Meng Hua;Qianbin Chen","doi":"10.1109/TNSE.2026.3653564","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3653564","url":null,"abstract":"Recently, reconfigurable intelligent surface (RIS) has been widely used to enhance the performance of millimeter wave (mmWave) communication systems, making beam alignment more challenging. To ensure efficient communication, this paper proposes a novel intelligent angle map-based beam alignment scheme for both general user equipments (UEs) and RIS-aided UEs simultaneously in a fast and effective way. Specifically, we construct a beam alignment architecture that utilizes only angular information. To obtain the angle information, the currently hottest seq2seq model – the Transformer – is introduced to offline learn the relationship between UE geographic location and the corresponding optimal beam direction. Based on the powerful machine learning model, the location-angle mapping function, i.e., the angle map, can be built. As long as the location information of UEs is available, the angle map can make the acquisition of beam alignment angles effortless. In the simulation, we utilize a ray-tracing-based dataset to verify the performance of the proposed scheme. It is demonstrated that the proposed scheme can achieve high-precision beam alignment and remarkable system performance without any beam scanning.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5833-5850"},"PeriodicalIF":7.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026387","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-13DOI: 10.1109/TNSE.2026.3653651
Dongyan Sui;Yufei Liu;Siyang Leng
Collective decision-making in networked systems is often shaped not only by peer interactions but also by persistent external influences. This paper introduces an intervened non-Bayesian social learning model thatexplicitly incorporates external information sources—whose beliefs remain fixed and potentially biased—into the belief-update process of a distributed multi-agent network. Analytical characterization of theproposed model reveals that such interventions disrupt strong consensus on the underlying true state, resulting in steady-state belief distributions that exhibit persistent oscillation and even polarization, consistent with empirical social observations. Building upon these insights, we propose a socially inspired adaptive algorithm for distributed online inference, which mitigates the rigidity of traditional non-Bayesian social learning updates and enables agents to remain responsive to environmental changes. Theoretical analysis and numerical experiments demonstrate that the proposed framework achieves enhanced adaptability and accurate online inference while preserving the decentralized cooperation mechanism of non-Bayesian social learning.
{"title":"Socially Inspired Adaptive Framework for Distributed Online Inference","authors":"Dongyan Sui;Yufei Liu;Siyang Leng","doi":"10.1109/TNSE.2026.3653651","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3653651","url":null,"abstract":"Collective decision-making in networked systems is often shaped not only by peer interactions but also by persistent external influences. This paper introduces an intervened non-Bayesian social learning model thatexplicitly incorporates external information sources—whose beliefs remain fixed and potentially biased—into the belief-update process of a distributed multi-agent network. Analytical characterization of theproposed model reveals that such interventions disrupt strong consensus on the underlying true state, resulting in steady-state belief distributions that exhibit persistent oscillation and even polarization, consistent with empirical social observations. Building upon these insights, we propose a socially inspired adaptive algorithm for distributed online inference, which mitigates the rigidity of traditional non-Bayesian social learning updates and enables agents to remain responsive to environmental changes. Theoretical analysis and numerical experiments demonstrate that the proposed framework achieves enhanced adaptability and accurate online inference while preserving the decentralized cooperation mechanism of non-Bayesian social learning.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5924-5942"},"PeriodicalIF":7.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082005","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/TNSE.2026.3651560
Hang Yin;Heli Zhang;Jianchi Zhu;Shan Yang;Jianxiu Wang;Peng Chen;Nan Ma
With the rising popularity of mobile video streaming, dynamic adaptive streaming over HTTP (DASH)-based bitrate adaptation has garnered significant attention in recent years. Existing studies in this area typically define a unified user quality of experience (QoE) by weighting various metrics, overlooking individual user preferences for QoE. Although some research has considered user preferences, it primarily focuses on optimizing bitrate selection alone, neglecting the joint allocation of communication resources that are tightly coupled with bitrate. In this paper, we propose QoE Optimization Enabler based on User Preference (QoEUP), a scheme for mobile video streaming, which dynamically adjusts bitrate, transmission power, and bandwidth based on channel quality and user preferences during mobility. The proposed scheme begins with training a reference model using deep reinforcement learning without incorporating user preferences. We then develop a user-friendly approach to collect user preferences and create a preference dataset. Finally, leveraging this dataset, we apply advanced direct preference optimization (DPO) to fine-tune the baseline model through supervised learning, effectively integrating individual QoE preferences. Simulation results demonstrate that QoEUP effectively aligns users' actual viewing experiences with their preferences in terms of video quality, playback smoothness, and device energy consumption.
{"title":"QoEUP: A Preference-Based QoE Optimization Scheme Using Human Feedback for Mobile Video Streaming","authors":"Hang Yin;Heli Zhang;Jianchi Zhu;Shan Yang;Jianxiu Wang;Peng Chen;Nan Ma","doi":"10.1109/TNSE.2026.3651560","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3651560","url":null,"abstract":"With the rising popularity of mobile video streaming, dynamic adaptive streaming over HTTP (DASH)-based bitrate adaptation has garnered significant attention in recent years. Existing studies in this area typically define a unified user quality of experience (QoE) by weighting various metrics, overlooking individual user preferences for QoE. Although some research has considered user preferences, it primarily focuses on optimizing bitrate selection alone, neglecting the joint allocation of communication resources that are tightly coupled with bitrate. In this paper, we propose QoE Optimization Enabler based on User Preference (QoEUP), a scheme for mobile video streaming, which dynamically adjusts bitrate, transmission power, and bandwidth based on channel quality and user preferences during mobility. The proposed scheme begins with training a reference model using deep reinforcement learning without incorporating user preferences. We then develop a user-friendly approach to collect user preferences and create a preference dataset. Finally, leveraging this dataset, we apply advanced direct preference optimization (DPO) to fine-tune the baseline model through supervised learning, effectively integrating individual QoE preferences. Simulation results demonstrate that QoEUP effectively aligns users' actual viewing experiences with their preferences in terms of video quality, playback smoothness, and device energy consumption.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"6156-6173"},"PeriodicalIF":7.9,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082023","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/TNSE.2026.3651664
Chenchen Fan;Qingling Wang;Shulong Zhao
Air–ground integrated mobile edge computing (AGI-MEC) integrates aerial and terrestrial resources to provide efficient computing services for massive mobile terminals (MTs), enabling 6G network intelligence. However, jointly optimizing offloading and power control remains challenging due to dynamic channels, resource constraints, communication overhead, and privacy concerns. To address these issues, this paper proposes a clustering-based two-layer federated deep reinforcement learning (CTL-FDRL) algorithm. A clustered federated training framework with parameter sharing is first developed, in which raw data exchange is replaced by periodic model aggregation, enabling MTs to learn offloading policies locally while preserving privacy. Furthermore, an efficient representative MT selection method is introduced. A self-organizing map (SOM)-based MT clustering method is designed to adaptively group MTs without a predefined number of clusters. Guided by the derived convergence bound, representative selection within each cluster is posed as a linear programming problem. The resulting representatives are used for federated aggregation, which reduces communication overhead without degrading model performance. Simulation results verify the superiority of CTL-FDRL, achieving about 28.7% higher cumulative reward, 31.5% lower delay, and 18.3% lower energy consumption compared with baseline algorithms.
{"title":"Communication-Efficient Federated Reinforcement Learning for Edge Offloading in AGI-MEC Systems","authors":"Chenchen Fan;Qingling Wang;Shulong Zhao","doi":"10.1109/TNSE.2026.3651664","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3651664","url":null,"abstract":"Air–ground integrated mobile edge computing (AGI-MEC) integrates aerial and terrestrial resources to provide efficient computing services for massive mobile terminals (MTs), enabling 6G network intelligence. However, jointly optimizing offloading and power control remains challenging due to dynamic channels, resource constraints, communication overhead, and privacy concerns. To address these issues, this paper proposes a clustering-based two-layer federated deep reinforcement learning (CTL-FDRL) algorithm. A clustered federated training framework with parameter sharing is first developed, in which raw data exchange is replaced by periodic model aggregation, enabling MTs to learn offloading policies locally while preserving privacy. Furthermore, an efficient representative MT selection method is introduced. A self-organizing map (SOM)-based MT clustering method is designed to adaptively group MTs without a predefined number of clusters. Guided by the derived convergence bound, representative selection within each cluster is posed as a linear programming problem. The resulting representatives are used for federated aggregation, which reduces communication overhead without degrading model performance. Simulation results verify the superiority of CTL-FDRL, achieving about 28.7% higher cumulative reward, 31.5% lower delay, and 18.3% lower energy consumption compared with baseline algorithms.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5622-5637"},"PeriodicalIF":7.9,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026383","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Federated Learning (FL) enables collaborative model training across decentralized edge devices while preserving data privacy. However, conventional FL frameworks face challenges due to data heterogeneity (non-IID), which impedes convergence, increases energy consumption and latency, and exposes the system to the data poisoning attack. To address the limitations, we propose a novel FL framework termed Federated Learning with Distributed Serial Pipeline Training (FedDSPT). On the client side, FedDSPT employs a dual grouping mechanism that organizes the edge devices into collaborative pipelines based on the feature similarity and diversity. The mechanism promotes more homogeneous intra-pipeline data distributions that approximate IID conditions, improving convergence behavior and reducing resource overhead. To optimize the pipeline formation, we apply the Held-Karp algorithm to determine minimal-cost, non-cyclic communication paths among the device groups. During the SPT phase, we incorporate the adversarial training through controlled injection of noise, enhancing robustness against the data poisoning attack arising from heterogeneous or malicious data sources. On the server side, FedDSPT utilizes an asynchronous pipeline-terminal model updates combined with the buffered aggregation technique to ensure timely and efficient global model synchronization. Experiments show that FedDSPT reduces energy consumption by 31.2% and training time by 26.7%, while demonstrating strong robustness and scalability under the large-scale deployments.
{"title":"FedDSPT: A Cost-Efficient and Low-Latency Federated Learning Framework Over Non-IID Data","authors":"Xumin Huang;Zican Huang;Weifeng Zhong;Maoqiang Wu;Ming Li;Shengli Xie","doi":"10.1109/TNSE.2026.3652982","DOIUrl":"https://doi.org/10.1109/TNSE.2026.3652982","url":null,"abstract":"Federated Learning (FL) enables collaborative model training across decentralized edge devices while preserving data privacy. However, conventional FL frameworks face challenges due to data heterogeneity (non-IID), which impedes convergence, increases energy consumption and latency, and exposes the system to the data poisoning attack. To address the limitations, we propose a novel FL framework termed Federated Learning with Distributed Serial Pipeline Training (FedDSPT). On the client side, FedDSPT employs a dual grouping mechanism that organizes the edge devices into collaborative pipelines based on the feature similarity and diversity. The mechanism promotes more homogeneous intra-pipeline data distributions that approximate IID conditions, improving convergence behavior and reducing resource overhead. To optimize the pipeline formation, we apply the Held-Karp algorithm to determine minimal-cost, non-cyclic communication paths among the device groups. During the SPT phase, we incorporate the adversarial training through controlled injection of noise, enhancing robustness against the data poisoning attack arising from heterogeneous or malicious data sources. On the server side, FedDSPT utilizes an asynchronous pipeline-terminal model updates combined with the buffered aggregation technique to ensure timely and efficient global model synchronization. Experiments show that FedDSPT reduces energy consumption by 31.2% and training time by 26.7%, while demonstrating strong robustness and scalability under the large-scale deployments.","PeriodicalId":54229,"journal":{"name":"IEEE Transactions on Network Science and Engineering","volume":"13 ","pages":"5870-5887"},"PeriodicalIF":7.9,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026420","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：