Pub Date : 2025-07-07DOI: 10.1109/LNET.2025.3586347
Tonghui Liang;Meng Hu;Enchang Sun
Federated learning is a distributed machine learning paradigm designed to facilitate collaborative model training while preserving user data privacy. However, in practical scenarios, data and model heterogeneity have emerged as significant barriers to its widespread application. Existing model heterogeneous personalized federated learning (MHPFL) algorithms often face trade-offs between performance and cost of computing and communication. To address this, we propose FedMoEKD, a novel MHPFL algorithm based on the Mixture of Experts (MoE) framework with a dynamic routing mechanism and decoupled training strategy. This approach decouples local MoE training and enhances the specialization capabilities of local expert models through knowledge distillation. By focusing on expert specialization, it achieves superior performance compared to state-of-the-art methods while maintaining minimal computational and communication overheads, including at least 0.58% higher accuracy on CIFAR-10 and 58% lower computation overheads than FedProto. Extensive experiments conducted on multiple datasets demonstrate that FedMoEKD outperforms existing MHPFL approaches in both convergence speed and model accuracy.
{"title":"Mixture of Specialized Experts for Model-Heterogeneous Personalized Federated Learning","authors":"Tonghui Liang;Meng Hu;Enchang Sun","doi":"10.1109/LNET.2025.3586347","DOIUrl":"https://doi.org/10.1109/LNET.2025.3586347","url":null,"abstract":"Federated learning is a distributed machine learning paradigm designed to facilitate collaborative model training while preserving user data privacy. However, in practical scenarios, data and model heterogeneity have emerged as significant barriers to its widespread application. Existing model heterogeneous personalized federated learning (MHPFL) algorithms often face trade-offs between performance and cost of computing and communication. To address this, we propose FedMoEKD, a novel MHPFL algorithm based on the Mixture of Experts (MoE) framework with a dynamic routing mechanism and decoupled training strategy. This approach decouples local MoE training and enhances the specialization capabilities of local expert models through knowledge distillation. By focusing on expert specialization, it achieves superior performance compared to state-of-the-art methods while maintaining minimal computational and communication overheads, including at least 0.58% higher accuracy on CIFAR-10 and 58% lower computation overheads than FedProto. Extensive experiments conducted on multiple datasets demonstrate that FedMoEKD outperforms existing MHPFL approaches in both convergence speed and model accuracy.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"224-228"},"PeriodicalIF":0.0,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145351911","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-02DOI: 10.1109/LNET.2025.3584929
Jie Yang;Junling Shi;Yunhe Sun;Aihua Men
Satellite communication often faces challenges like path loss and high latency when supplementing terrestrial networks. We propose high-altitude platforms (HAPs) as relay nodes constructing a satellite-HAP-terrestrial integrated network (SHTIN) architecture to mitigate these issues. This architecture enhances link stability and provides low-latency computation offloading. We design a prediction-based task planning (TP) algorithm that enables efficient offload node selection by predicting resource demands, preventing overload or resource waste of agents. Guided by the TP algorithm, we employ a multi-agent proximal policy optimization (MAPPO) algorithm to optimize offloading policies, reducing energy consumption while maximizing resource utilization within tolerable task delays. Simulations confirm that our method outperforms existing approaches in resource efficiency and energy consumption, achieving superior overall performance.
{"title":"Task Prediction-Based Edge Computing Offloading of Satellite-HAP-Terrestrial Integrated Network","authors":"Jie Yang;Junling Shi;Yunhe Sun;Aihua Men","doi":"10.1109/LNET.2025.3584929","DOIUrl":"https://doi.org/10.1109/LNET.2025.3584929","url":null,"abstract":"Satellite communication often faces challenges like path loss and high latency when supplementing terrestrial networks. We propose high-altitude platforms (HAPs) as relay nodes constructing a satellite-HAP-terrestrial integrated network (SHTIN) architecture to mitigate these issues. This architecture enhances link stability and provides low-latency computation offloading. We design a prediction-based task planning (TP) algorithm that enables efficient offload node selection by predicting resource demands, preventing overload or resource waste of agents. Guided by the TP algorithm, we employ a multi-agent proximal policy optimization (MAPPO) algorithm to optimize offloading policies, reducing energy consumption while maximizing resource utilization within tolerable task delays. Simulations confirm that our method outperforms existing approaches in resource efficiency and energy consumption, achieving superior overall performance.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"185-189"},"PeriodicalIF":0.0,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145352145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01DOI: 10.1109/LNET.2025.3584344
Yunpeng Zhou;Gaolei Fei;Jian Ye;Xuemeng Zhai;Guangmin Hu
This letter presents a knowledge inference-based (KI) method for topology construction. KI builds a multi-source heterogeneous network knowledge graph, integrating multi-modal data, including IP interface-level topology, geospatial information, and autonomous system affiliations, to achieve a semantic representation of network topology. A hierarchical-aware embedding model (HAKE) is utilized to extract semantic features of topology entities and encode them as polar coordinate vectors. By incorporating routing protocol constraints, the tasks of IP alias resolution and anonymous router identification are reformulated as a constrained maximum likelihood estimation problem, enabling joint inference and completion of topology relationships. Experimental results demonstrate that the proposed method significantly enhances topology reconstruction accuracy across various scenarios with different anonymous interface ratios and probing intensities while maintaining robust performance even in high anonymous interface ratio conditions.
{"title":"Large-Scale Network Router-Level Topology Construction Based on Knowledge Inference","authors":"Yunpeng Zhou;Gaolei Fei;Jian Ye;Xuemeng Zhai;Guangmin Hu","doi":"10.1109/LNET.2025.3584344","DOIUrl":"https://doi.org/10.1109/LNET.2025.3584344","url":null,"abstract":"This letter presents a knowledge inference-based (KI) method for topology construction. KI builds a multi-source heterogeneous network knowledge graph, integrating multi-modal data, including IP interface-level topology, geospatial information, and autonomous system affiliations, to achieve a semantic representation of network topology. A hierarchical-aware embedding model (HAKE) is utilized to extract semantic features of topology entities and encode them as polar coordinate vectors. By incorporating routing protocol constraints, the tasks of IP alias resolution and anonymous router identification are reformulated as a constrained maximum likelihood estimation problem, enabling joint inference and completion of topology relationships. Experimental results demonstrate that the proposed method significantly enhances topology reconstruction accuracy across various scenarios with different anonymous interface ratios and probing intensities while maintaining robust performance even in high anonymous interface ratio conditions.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"210-214"},"PeriodicalIF":0.0,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145352052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
By leveraging data-centric forwarding and in-network caching, Vehicular Named Data Networking (VNDN) enhances communication in vehicular networks. However, advanced techniques are still required to support effective data retrieval while handling error-prone wireless channels, short-lived contacts and intermittent connectivity. In this letter, we introduce a mechanism based on the Chinese Remainder Theorem (CRT), which fragments content into smaller redundant pieces, enabling efficient content caching and retrieval even in lossy environments. Simulation results in a highway scenario demonstrate improved data accessibility, scalability, and reduced retrieval delays, highlighting the potential of CRT to enhance VNDN performance under dynamic conditions.
{"title":"Improving Content Retrieval in Vehicular NDN Using the Chinese Remainder Theorem","authors":"Marica Amadeo;Filippo Battaglia;Giuseppe Campobello","doi":"10.1109/LNET.2025.3584200","DOIUrl":"https://doi.org/10.1109/LNET.2025.3584200","url":null,"abstract":"By leveraging data-centric forwarding and in-network caching, Vehicular Named Data Networking (VNDN) enhances communication in vehicular networks. However, advanced techniques are still required to support effective data retrieval while handling error-prone wireless channels, short-lived contacts and intermittent connectivity. In this letter, we introduce a mechanism based on the Chinese Remainder Theorem (CRT), which fragments content into smaller redundant pieces, enabling efficient content caching and retrieval even in lossy environments. Simulation results in a highway scenario demonstrate improved data accessibility, scalability, and reduced retrieval delays, highlighting the potential of CRT to enhance VNDN performance under dynamic conditions.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"190-194"},"PeriodicalIF":0.0,"publicationDate":"2025-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145352042","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
When constructing a vehicular network, safety, efficiency and reliability represent the fundamental requirement. The usually encountered Long Range (LoRa) communication protocol has overcome the coverage limitations while focusing on the trade-offs between dependable transmissions and resource expenditure. Such a system supports battery-powered edge devices to monitor the coverage area. Through the LoRa Wide Area Network (LoRaWAN) architecture, a more flexible and locality-driven network can be structured. This letter establishes a broadcast-based discovery system for a vehicular environment to simulate potential imminent dangers. This letter examines a crisis management system that provides detailed insights into the resource intensity associated with non-discrete events. The research primarily focuses on optimizing network capacity and effectively disseminating critical information to those affected by events involving vehicles.
在构建车载网络时，安全、高效、可靠是最基本的要求。通常遇到的远程（LoRa）通信协议克服了覆盖限制，同时专注于可靠传输和资源消耗之间的权衡。这种系统支持电池供电的边缘设备来监控覆盖区域。通过LoRaWAN （LoRa Wide Area Network）架构，可以构建一个更加灵活和位置驱动的网络。这封信为车辆环境建立了一个基于广播的发现系统，以模拟潜在的迫在眉睫的危险。这封信考察了一个危机管理系统，该系统提供了与非离散事件相关的资源强度的详细见解。研究的重点是优化网络容量，并有效地向受车辆事件影响的人员传播关键信息。
{"title":"LoRaWAN-Based Emergency Framework for Vehicular Networks","authors":"Dimitrios Georgiadis;Konstantina Karathanasopoulou;Eirini Liotou;George Dimitrakopoulos","doi":"10.1109/LNET.2025.3583492","DOIUrl":"https://doi.org/10.1109/LNET.2025.3583492","url":null,"abstract":"When constructing a vehicular network, safety, efficiency and reliability represent the fundamental requirement. The usually encountered Long Range (LoRa) communication protocol has overcome the coverage limitations while focusing on the trade-offs between dependable transmissions and resource expenditure. Such a system supports battery-powered edge devices to monitor the coverage area. Through the LoRa Wide Area Network (LoRaWAN) architecture, a more flexible and locality-driven network can be structured. This letter establishes a broadcast-based discovery system for a vehicular environment to simulate potential imminent dangers. This letter examines a crisis management system that provides detailed insights into the resource intensity associated with non-discrete events. The research primarily focuses on optimizing network capacity and effectively disseminating critical information to those affected by events involving vehicles.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"205-209"},"PeriodicalIF":0.0,"publicationDate":"2025-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145351926","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-13DOI: 10.1109/LNET.2025.3579358
Mingchuang Zhang;Hongbo Tang;Jie Yang;Hang Qiu;Yu Zhao;Mingyan Xu;Yi Bai
The Network Data Analytics Function (NWDAF) proposed by 3GPP provides a novel solution for anomaly traffic detection in 5G core networks (5GC). However, existing studies generally adopt single model, which struggle to effectively handle data from different network functions (NFs). To address this problem, this letter proposes a Self-Adaptive Multi-Model Fusion (SAMMF) framework for NWDAF, which can processes the different NFs data. The SAMMF framework consists of four core modules: the data collection module, which is responsible for data statistics and collection; the data preprocessing module, focusing on data cleaning and feature engineering; the self-adaptive multi-model training module, which selects high-performance models from the model library using an adaptive threshold algorithm; and the multi-model fusion module, which fuses the results of the selected models to derive the final result. We evaluated the SAMMF using two commonly used network anomaly detection datasets. Experimental results show that, compared to existing baseline methods, SAMMF demonstrates significant advantages in handling different NFs data, providing a superior solution for anomaly traffic detection in 5GC.
{"title":"SAMMF: A Self-Adaptive Multi-Model Fusion Framework for NWDAF Traffic Anomaly Detection","authors":"Mingchuang Zhang;Hongbo Tang;Jie Yang;Hang Qiu;Yu Zhao;Mingyan Xu;Yi Bai","doi":"10.1109/LNET.2025.3579358","DOIUrl":"https://doi.org/10.1109/LNET.2025.3579358","url":null,"abstract":"The Network Data Analytics Function (NWDAF) proposed by 3GPP provides a novel solution for anomaly traffic detection in 5G core networks (5GC). However, existing studies generally adopt single model, which struggle to effectively handle data from different network functions (NFs). To address this problem, this letter proposes a Self-Adaptive Multi-Model Fusion (SAMMF) framework for NWDAF, which can processes the different NFs data. The SAMMF framework consists of four core modules: the data collection module, which is responsible for data statistics and collection; the data preprocessing module, focusing on data cleaning and feature engineering; the self-adaptive multi-model training module, which selects high-performance models from the model library using an adaptive threshold algorithm; and the multi-model fusion module, which fuses the results of the selected models to derive the final result. We evaluated the SAMMF using two commonly used network anomaly detection datasets. Experimental results show that, compared to existing baseline methods, SAMMF demonstrates significant advantages in handling different NFs data, providing a superior solution for anomaly traffic detection in 5GC.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"239-243"},"PeriodicalIF":0.0,"publicationDate":"2025-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145352053","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-13DOI: 10.1109/LNET.2025.3579672
Shoya Imanaka;Masahiro Inoue;Eiji Oki
This letter proposes a distributed server allocation model with single-database connectivity for delay-sensitive Internet-of-Things monitoring services. Based on preventive start-time optimization against server failure, it minimizes the maximum transmission delay across all failure scenarios. Unlike the conventional two-database connectivity model, which restricts application server placement, the proposed model allows connection to a single database server, expanding allocation flexibility. Numerical results show that the proposed model reduces the maximum delay under server failure by up to 10.7% in our examined cases compared to the conventional model.
{"title":"Failure-Resilient Server Allocation Model With Single-Database Connectivity for Delay-Sensitive Internet-of-Things Monitoring","authors":"Shoya Imanaka;Masahiro Inoue;Eiji Oki","doi":"10.1109/LNET.2025.3579672","DOIUrl":"https://doi.org/10.1109/LNET.2025.3579672","url":null,"abstract":"This letter proposes a distributed server allocation model with single-database connectivity for delay-sensitive Internet-of-Things monitoring services. Based on preventive start-time optimization against server failure, it minimizes the maximum transmission delay across all failure scenarios. Unlike the conventional two-database connectivity model, which restricts application server placement, the proposed model allows connection to a single database server, expanding allocation flexibility. Numerical results show that the proposed model reduces the maximum delay under server failure by up to 10.7% in our examined cases compared to the conventional model.","PeriodicalId":100628,"journal":{"name":"IEEE Networking Letters","volume":"7 3","pages":"215-219"},"PeriodicalIF":0.0,"publicationDate":"2025-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11036332","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145352043","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","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：