Computer Networks最新文献

{"title":"GraphVeri: A NAR-based control plane verification framework for routing protocols","authors":"Shangsen Li ,&nbsp;Lailong Luo ,&nbsp;Changhao Qiu ,&nbsp;Bangbang Ren ,&nbsp;Yun Zhou ,&nbsp;Deke Guo ,&nbsp;Richard T.B. Ma","doi":"10.1016/j.comnet.2025.111635","DOIUrl":"10.1016/j.comnet.2025.111635","url":null,"abstract":"<div><div>The distributed control plane routing protocols of networks are inherently complex and prone to configured incorrectly, such as BGP and OSPF, necessitating rigorous verification to ensure that configurations meet requirements. The classic methods for configuration verification predominantly rely on formal verification techniques, which model the intricate relationships among network configurations, protocols and the corresponding forwarding behaviors, under some assumptions of the network environment. However, these methods are lack of scalability (the verification time increases exponentially as topology scales) and generality (requiring substantial manual effort for development and maintenance). This paper introduces a novel neural algorithmic reasoning (NAR) based verification framework called GraphVeri, aiming at distributed routing protocol configuration verification. Our approach can learn how to verify from the perfect mapping from configurations to specification satisfactions directly and continuously, thereby capturing the underlying knowledge of distributed control plane protocols and their verification processes. With such a learning-based verifier, we can achieve comprehensive end-to-end verification with perfect scalability and extendability, and without the burdensome task of formal modeling typically associated with distributed routing protocols. Furthermore, the inductive learning capability of GraphVeri enables the verifier to learn how to integrate the local node attribute information to generate embeddings for previously unseen nodes. Evaluations conducted on the Topology Zoo dataset and BGP&amp;OSPF protocols demonstrate that our NAR-based learning verifiers attain high accuracy, efficiency and scalability. GraphVeri achieves comparable accuracy to GraphGAT, which was initially developed for network synthesis, while at <span><math><mrow><mn>2</mn><mo>×</mo></mrow></math></span> (GPU) and <span><math><mrow><mn>10</mn><mo>×</mo></mrow></math></span> (CPU) speed up. Compared with the classic verifiers, GraphVeri (CPU) can attain a speed up of 2.93–38.28<span><math><mo>×</mo></math></span> and 2300–12764<span><math><mo>×</mo></math></span> to Batfish and Minesweeper respectively; GraphVeri (GPU) attain a speed of 33.51–366.29<span><math><mo>×</mo></math></span> and 30434–217653<span><math><mo>×</mo></math></span> to Batfish and Minesweeper respectively. Moreover, the verification time of GraphVeri increases slower than that of the classic verifiers.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111635"},"PeriodicalIF":4.6,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144878778","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}

{"title":"secPEFL: Strengthening federated learning security for Portable Executable malware detection in distributed networks","authors":"Trinh Gia Huy,&nbsp;Luong Nguyen Thanh Nhan,&nbsp;Nguyen Tan Cam","doi":"10.1016/j.comnet.2025.111638","DOIUrl":"10.1016/j.comnet.2025.111638","url":null,"abstract":"<div><div>Traditional centralized malware detection approaches are increasingly vulnerable to privacy risks and data breaches, particularly given stringent regulatory requirements. To address these challenges, we propose a Federated learning-based system for malware classification on PE executable files, emphasizing enhanced data privacy and security. Our approach leverages a Convolutional Neural Network architecture with two modules: a detection module for detecting malicious files and a classification module for identifying malware types and supporting defense strategies. The system operates on grayscale images and incorporates advanced security measures, including Secure Sockets Layer for secure communication, InterPlanetary File System for distributed storage, and Local Differential Privacy to counter inference attacks. The proposed system mitigates Sybil attacks through a participant selection mechanism based on reputation history stored on the blockchain network. The blockchain is also used as a reward platform for contributors, utilizing a Shapley value-based reward mechanism from game theory. Experimental results show that the proposed system delivers superior malware classification performance while maintaining security aspects. The highest accuracy achieved is 96.32% on IID (Independent and Identically Distributed) data and 88.06% on non-IID data for malware classification tasks. The experiments also reveal that as the level of noise added using Differential Privacy increases, security improves, but the model’s accuracy decreases correspondingly.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111638"},"PeriodicalIF":4.6,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886586","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}

{"title":"Maximizing computation rate for NOMA-based WPT-MEC with user cooperation under nonlinear EH model","authors":"Yunge Duan,&nbsp;Zhenbo Liu,&nbsp;Shuang Fu","doi":"10.1016/j.comnet.2025.111639","DOIUrl":"10.1016/j.comnet.2025.111639","url":null,"abstract":"<div><div>Over the past few years, the market size of the Internet of Things (IoT) has achieved a leapfrog growth. However, the large-scale data processing and the limited battery capacity of IoT devices have posed severe challenges to the existing IoT. To address this issue, mobile edge computing (MEC) technology and wireless power transfer (WPT) technology have been introduced, significantly reducing the computing pressure on IoT devices and achieving sustainable energy supply. To optimize the computing capacity of the IoT, this paper studies a wirelessly powered transmission mobile edge computing (WPT-MEC) system assisted by user cooperation (UC) and non-orthogonal multiple access (NOMA). A nonlinear energy harvesting (EH) model that is more in line with the actual energy collection process in reality has been adopted. By jointly optimizing the time allocation, transmission power, and power distribution across multiple energy beams at the HAP, the system’s weighted sum computation rate (WSCR) is maximized. To solve the proposed non-convex optimization problem, an iterative optimization algorithm based on successive convex approximation (SCA) is designed. Firstly, the non-convex optimization problem is reformulated using the variable substitution method to transform it into a more manageable form, and then SCA is applied to solve it. Numerical results illustrate that the proposed scheme can improve the overall performance of the system compared with the traditional benchmark schemes.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111639"},"PeriodicalIF":4.6,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144893222","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}

{"title":"ZETROS: A zero-trust IoT network security framework using distributed blacklisting, trust scoring and smart contracts","authors":"Cem Ata Baykara ,&nbsp;Ilgın Şafak ,&nbsp;Kubra Kalkan","doi":"10.1016/j.comnet.2025.111601","DOIUrl":"10.1016/j.comnet.2025.111601","url":null,"abstract":"<div><div>The purpose of Internet of Things (IoT) security is to ensure the availability, confidentiality, and integrity of IoT networks. However, due to the heterogeneity of IoT devices and the possibility of attacks of various kinds from both inside and outside the network, securing an IoT network is a difficult task. Handshake protocols are useful for achieving mutual authentication, which allows secure inclusion of devices into the network. By verifying that the information they receive is accurate and from a trusted source, mutual authentication minimizes the possibility that a malicious actor will compromise their connections. However, handshake protocols do not protect devices from attackers in the network. Use of autonomous anomaly detection and blacklisting prevents nodes with anomalous behavior from joining, re-joining, or remaining in the network. Similarly, trust scoring is another popular method that can be used to increase the resilience of the network against trust based system attacks. In view of the above, the contributions of this paper are three-fold. First, to ensure the security of the IoT network from outsider attacks in a zero-trust environment, we propose a new handshake protocol based on Physical Unclonable Functions that can be used in IoT device discovery and mutual authentication between the IoT device and the server. The proposed protocol is resilient to Man-in-the-Middle, replay and forgery attacks, as proven in our security analysis. Secondly, we propose a real-time intrusion and anomaly detection framework based on machine learning to prevent network-based attacks from insiders. Finally, we propose a trust system which utilizes feedback mechanisms based on smart contracts for managing the trust of a dynamic IoT network to increase resilience against behavioral attacks. Simulation results show that by using blacklisting, our trust management model provides greater resilience against trust-based attacks compared to similar blockchain-based trust models in the literature, and the proposed distributed IoT network security framework can secure an IoT network from both internal and external attacks, even in an environment where half of the devices in the network are compromised.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111601"},"PeriodicalIF":4.6,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144908917","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}

{"title":"Sustainable 6G architecture: An organic evolution of 5G networks","authors":"Özgür Umut Akgül ,&nbsp;Antonio Varvara ,&nbsp;Antonio de la Oliva ,&nbsp;Panagiotis Charatsaris ,&nbsp;Maria Diamanti ,&nbsp;Pere Garau Burguera ,&nbsp;Mårten Ericson ,&nbsp;Stefan Wänstedt ,&nbsp;Marcin Ziółkowski ,&nbsp;Halina Tarasiuk ,&nbsp;Hamed Hellaoui ,&nbsp;Symeon Papavassiliou ,&nbsp;Vasileios Tsekenis ,&nbsp;Sokratis Barmpounakis ,&nbsp;Panagiotis Demestichas ,&nbsp;Bahare M. Khorsandi ,&nbsp;Hasanin Harkous","doi":"10.1016/j.comnet.2025.111629","DOIUrl":"10.1016/j.comnet.2025.111629","url":null,"abstract":"<div><div>The evolution of cellular networks towards 6G presents an opportunity to enhance not only coverage and capacity but also environmental, social, and economic sustainability. This paper proposes a sustainable cloud-native and modular 6G architecture that integrates AI-driven resource orchestration and intent-based network management. The proposed design introduces a modular approach where network functions are structured into adaptable building blocks, allowing for seamless scalability and flexible deployment tailored to emerging use cases. AI is embedded across multiple layers to optimize energy efficiency, improve service accessibility, and enable intelligent decision-making. Furthermore, the architecture leverages integrated network-compute paradigms, enhancing service placement efficiency and minimizing energy consumption through optimized edge computing solutions. Intent-based management further refines network operations by shifting from manual configurations to high-level policy abstractions, improving resource efficiency while reducing operational complexity. This paper also explores enhancements to both the Radio Access Network (RAN) and Core Network (CN), improving network modularization and signaling efficiency. The proposed framework ensures a smooth transition from 5G to 6G while addressing key challenges such as network sustainability, trustworthiness, and digital inclusivity. By incorporating these elements, the envisioned 6G architecture aims to establish a resilient, energy-efficient, and future-proof network ecosystem.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111629"},"PeriodicalIF":4.6,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144878788","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}

{"title":"A survey on 5G private and B5G network threats and safeguarding AI-based security mechanisms through the layered analysis","authors":"Saman Tariq ,&nbsp;Eva Rodriguez Luna ,&nbsp;Xavier Masip Bruin ,&nbsp;Rodrigo Diaz ,&nbsp;Josep Martrat ,&nbsp;Panagiotis Trakadas","doi":"10.1016/j.comnet.2025.111594","DOIUrl":"10.1016/j.comnet.2025.111594","url":null,"abstract":"&lt;div&gt;&lt;div&gt;The fifth-generation (5G) mobile network has shifted the paradigm in connectivity, high-speed data transmission, ultra-low latency, ultra-high throughput, multisense transmission, and ultra-high reliability. Today, industries are increasingly adopting 5G private networks, which also handle sensitive data related to business information, trade secrets, and personal data. Attacks on 5G private networks can potentially result in losing a competitive advantage since different security threats and vulnerabilities progressively target these networks. The unique infrastructure of the 5G network architecture and key enabling technologies exposes them to various vulnerabilities that attackers can target to breach sensitive data, steal information, and disrupt critical systems. Therefore, paying special attention to the security issues of 5G private networks is essential. The advancement of future wireless technology came about because of the different and diverse nature of connected devices, in contrast to the previous generation of mobile networks. The B5G network provides support to open network platforms, open interfaces, and the integration of different key enabling technologies helps manage network services and deploy new services needed for diverse requirements. At the same time, it has increased the attack surface compared to previous-generation networks. Therefore, it is imperative to conduct a review that focuses on addressing and classifying the different emerging threats in the private 5G and B5G networks in a distinctive way. In this paper, we have adopted a layered architecture from an industrial use case of 5G private networks to identify and classify different threats in 5G private networks. The study also characterized and modeled the different threats using information on the type of attack, entry points, and impact of the attack on the architecture layer. Moreover, an analysis of key enablers in 5G private and B5G networks and information on security threats and cyber-attacks is also presented. To accommodate the emerging threats in next-generation wireless technology, we have classified and modeled the different threats in the B5G domain using the common involved layer, which includes perception, network, and application layers in Hexa-x E2E and 6G IoT-enabled architecture. Organizations and projects that do not actively engage in cutting-edge technologies will lose their competitive advantage in the evolving technological landscape. This study has mapped the identified threat categories with different types of threats that could occur at different layers and assessed the entry points of the attacker. We have also identified the attack’s impact at each layer using security requirements related to confidentiality, availability, and integrity. Furthermore, this study reviews different AI-enabled solutions that can add value in preserving the security of 5G and B5G networks. Through this review analysis, we can conclude that the ","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111594"},"PeriodicalIF":4.6,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144878787","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}

{"title":"SMUSAC: Lightweight federated learning framework for SUNETs with tolerance of data loss and node compromise","authors":"Wenhao Cheng ,&nbsp;Xia Feng ,&nbsp;Liangmin Wang ,&nbsp;Zhan Xie ,&nbsp;Liang Wang ,&nbsp;Siben Tian","doi":"10.1016/j.comnet.2025.111626","DOIUrl":"10.1016/j.comnet.2025.111626","url":null,"abstract":"<div><div>Satellite-Assisted Unmanned-System Networks (SUNETs) are emerging network applications that leverage satellites to support ubiquitous data-driven services, such as autonomous underwater vehicles and unmanned aircraft systems. In these applications, transmitting data over external networks poses a risk of privacy leakage. Usually, federated learning is used to prevent the direct leakage of raw data; however, its effectiveness and robustness in SUNETs are constrained due to two key challenges arising from limited bandwidth and unmanned nodes: (a) <em>data loss</em>, some nodes may fail to transmit data back to the server in time; (b) <em>node compromise</em>, unmanned nodes might be controlled by adversaries, even uploading malicious data to the server. To address these challenges, we propose a lightweight federated learning framework, called SMUSAC, which includes three stages: Sparsifying Model, Uploading Signs, and Aggregating with Compensation. Specifically, we design a sign-based updating mechanism for sparsified models, rather than transmitting model parameters or gradients over the communication link. It improves SMUSAC’s tolerance to data loss and node compromise by relying solely on the sign of updates rather than their specific values, while also reducing bandwidth demands. Additionally, an error-compensation mechanism is employed to mitigate the accuracy loss caused by sparsification. We theoretically analyze the convergence of SMUSAC under a non-convex cost function. Simulation results show that SMUSAC exhibits significant resilience under adverse conditions, maintaining stable performance even with 40% of nodes compromised, and outperforms seven baselines across multiple evaluation metrics.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111626"},"PeriodicalIF":4.6,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886487","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}

{"title":"Source identification for worm propagation: A graph neural network approach and evaluation on social network and internet datasets","authors":"Qitao Huo,&nbsp;Peng Zhou","doi":"10.1016/j.comnet.2025.111616","DOIUrl":"10.1016/j.comnet.2025.111616","url":null,"abstract":"<div><div>Source identification plays an essential role in the analysis and forensics of worm propagation but unfortunately is quite challenging to solve due to the limited traces and clues left on the observed propagation graphs. State-of-the-art solutions to source identification are mostly based on unsupervised graph induction and reasoning, hence missing the chances to find more trails from additional origins of information for worm tracing. In this paper, we go beyond unsupervised source identification and make perhaps the first attempt to design a supervised solution, to “borrow” outside information to facilitate the detection of the propagation sources. Our basic idea is to apply a graph neural network (GNN) to learn the additional clues (specifically the node state distributions over the graph structures) from a training set of propagation graph samples whose sources are known in advance, hence able to model the mapping relationship between the different node state distributions and the many different nodes as the sources for propagation. This way, we can wisely convert the unsupervised source identification problem to a supervised classification of propagation graphs with the sources as class labels, thereby tracing back the given worm later guided by the similar propagation behaviors found on the sampled propagation graphs. We understand that the direct use of the GNN model is not quite effective in the condition of large graphs since a large number of nodes should be considered individual class labels for classification and accordingly propose a hierarchical improvement. That is, we cluster the nodes from the large graph into several smaller subgraphs (i.e., communities) and then deploy a set of GNN models through a hierarchical architecture for these subgraphs, hence being able to largely reduce the number of class labels for each of the GNN models over this architecture. To evaluate the effectiveness of our solution, we have run extensive source identification experiments using the worm propagation graphs simulated from both the synthetic and social network and Internet datasets. Our results have successfully confirmed a higher identification accuracy (in terms of the length of the shortest path from the identified source to the true one) by our supervised solution compared with the competing counterparts. For the best case, we can improve the identification accuracy up to ten times the magnitude.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111616"},"PeriodicalIF":4.6,"publicationDate":"2025-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144861200","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}

{"title":"Maintaining source–destination connectivity in uncertain networks under adversarial attack","authors":"Jianzhi Tang ,&nbsp;Luoyi Fu ,&nbsp;Yu-E Sun ,&nbsp;Xinbing Wang ,&nbsp;He Huang","doi":"10.1016/j.comnet.2025.111607","DOIUrl":"10.1016/j.comnet.2025.111607","url":null,"abstract":"<div><div>This paper investigates the problem of maintaining the connectivity between two vertices, a source and a destination, in an uncertain network under adversarial attack, where a defender preserves crucial links to prevent the source–destination vertex pair from being disconnected by an attacker. In contrast with prior art that mostly focuses on the overall network connectivity, in this work connectivity maintenance is restricted to a pair of selected vertices, which may provide insights into reliable point-to-point connection. We model the network as a random graph where each link carries both an existence probability and a probing cost, and seek to design a defensive strategy that ensures source–destination connectivity under minimum probing expenditure, regardless of adversarial behavior. To this end, we first delve into the computational complexity of the problem by establishing its NP-hardness, and put forth an optimal defensive strategy leveraging dynamic programming. Due to the prohibitive price of attaining optimality, we further design two approximate defensive strategies aimed at pursuing effective defensive performance within polynomial time, in which the first one is a path-based heuristic strategy that iteratively extends a preserved path by probing links with high utility regarding source–destination connectivity, and the second one is a cut-based minimax strategy that prioritizes the links in the minimum potential source–destination cut in order to minimize the possible worst-case loss suffered by the defender with a constant approximation ratio. Extensive experiments conducted on synthetic and real-world network datasets under diverse attacking strategies validate the superiority of the proposed strategies in both effectiveness and robustness over baselines.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111607"},"PeriodicalIF":4.6,"publicationDate":"2025-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144878781","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}

{"title":"SCDFL: A Spectral Clustering-based framework for accelerating convergence in Decentralized Federated Learning","authors":"Faisal Alshami ,&nbsp;Lin Yao ,&nbsp;Xin Wang ,&nbsp;Guowei Wu","doi":"10.1016/j.comnet.2025.111615","DOIUrl":"10.1016/j.comnet.2025.111615","url":null,"abstract":"<div><div>Decentralized Federated Learning (DFL) is a popular distributed machine learning framework that facilitates collaboration among multiple clients without dependence on a central server to develop a global model. This architecture faces issues with client convergence, resulting in network congestion and slower convergence during the DFL process. These challenges stem from various communications topologies and the non-independent and non-identically distributed nature of data on terminal devices in real-world scenarios, which affect both model convergence speed and overall terminal performance. Therefore, we propose SCDFL, a federated learning framework that leverages spectral clustering to efficiently and scalably handle client data heterogeneity. SCDFL introduces a novel spectral clustering strategy that focuses on grouping clients based on their characteristics. Key components include reducing the dimensionality of the client data by incremental PCA, which includes high-dimensional model updates or feature vectors, making the clustering process more efficient. Then, a similarity matrix based on the reduced data will be computed to measure client similarity. Utilizing this matrix, we apply spectral clustering to group clients with similar data characteristics. Finally, we apply the aggregation in intra-cluster and inter-cluster to the updated global model. Extensive experiments have been conducted across different topologies, and the results demonstrate that SCDFL achieves higher accuracy, faster convergence, reduced communication overhead, and improved generalization, particularly on complex datasets like MNIST, CIFAR10, and CIFAR100, while efficiently handling data heterogeneity and optimizing resource utilization across various network topologies.</div></div>","PeriodicalId":50637,"journal":{"name":"Computer Networks","volume":"271 ","pages":"Article 111615"},"PeriodicalIF":4.6,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886488","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}