Software-defined wireless mesh networks are being increasingly deployed in diverse settings, such as smart cities and public Wi-Fi access infrastructures. The signal propagation and interference issues that typically characterize these environments can be handled by employing SDN controller mechanisms, effectively monitoring link quality and triggering appropriate mitigation strategies, such as adjusting link and/or routing protocols. In this paper, we propose an unsupervised machine learning (ML) online framework for link quality detection consisting of: 1) improved preprocessing clustering algorithm, based on elastic similarity measures, to efficiently characterize wireless links in terms of reliability, and 2) a novel change point (CP) detector for the real-time identification of anomalies in the quality of selected links, which minimizes the overestimation error through the incorporation of a rank-based test and a recursive max-type procedure. In this sense, considering the communication constraints of such environments, our approach minimizes the detection overhead and the inaccurate decisions caused by overestimation. The proposed detector is validated, both on its individual components and as an overall mechanism, against synthetic but also real data traces; the latter being extracted from real wireless mesh network deployments.
软件定义的无线网格网络正越来越多地部署在各种环境中，如智能城市和公共 Wi-Fi 接入基础设施。信号传播和干扰问题是这些环境的典型特征，可通过采用 SDN 控制器机制来处理，有效监控链路质量并触发适当的缓解策略，如调整链路和/或路由协议。在本文中，我们提出了一种用于链路质量检测的无监督机器学习（ML）在线框架，该框架由以下部分组成：1) 基于弹性相似度量的改进型预处理聚类算法，可有效描述无线链路的可靠性特征；以及 2) 用于实时识别选定链路质量异常的新型变化点（CP）检测器，该检测器通过基于等级的测试和递归最大值类型程序将高估误差降至最低。从这个意义上说，考虑到此类环境的通信限制，我们的方法最大限度地减少了检测开销和高估造成的不准确决策。我们对所提出的检测器进行了验证，无论是对其单个组件还是作为一个整体机制，都进行了合成和真实数据追踪；后者是从真实的无线网状网络部署中提取的。
{"title":"A Link-Quality Anomaly Detection Framework for Software-Defined Wireless Mesh Networks","authors":"Sotiris Skaperas;Lefteris Mamatas;Vassilis Tsaoussidis","doi":"10.1109/TMLCN.2024.3388973","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3388973","url":null,"abstract":"Software-defined wireless mesh networks are being increasingly deployed in diverse settings, such as smart cities and public Wi-Fi access infrastructures. The signal propagation and interference issues that typically characterize these environments can be handled by employing SDN controller mechanisms, effectively monitoring link quality and triggering appropriate mitigation strategies, such as adjusting link and/or routing protocols. In this paper, we propose an unsupervised machine learning (ML) online framework for link quality detection consisting of: 1) improved preprocessing clustering algorithm, based on elastic similarity measures, to efficiently characterize wireless links in terms of reliability, and 2) a novel change point (CP) detector for the real-time identification of anomalies in the quality of selected links, which minimizes the overestimation error through the incorporation of a rank-based test and a recursive max-type procedure. In this sense, considering the communication constraints of such environments, our approach minimizes the detection overhead and the inaccurate decisions caused by overestimation. The proposed detector is validated, both on its individual components and as an overall mechanism, against synthetic but also real data traces; the latter being extracted from real wireless mesh network deployments.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"495-510"},"PeriodicalIF":0.0,"publicationDate":"2024-04-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10499246","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140639411","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}
The advancement of deep learning (DL) techniques has led to significant progress in Automatic Modulation Classification (AMC). However, most existing DL-based AMC methods require massive training samples, which are difficult to obtain in non-cooperative scenarios. The identification of modulation types under small sample conditions has become an increasingly urgent problem. In this paper, we present a novel few-shot AMC model named the Spatial Temporal Transductive Modulation Classifier (STTMC), which comprises two modules: a feature extraction module and a graph network module. The former is responsible for extracting diverse features through a spatiotemporal parallel network, while the latter facilitates transductive decision-making through a graph network that uses a closed-form solution. Notably, STTMC classifies a group of test signals simultaneously to increase stability of few-shot model with an episode training strategy. Experimental results on the RadioML.2018.01A and RadioML.2016.10A datasets demonstrate that the proposed method perform well in 3way-Kshot, 5way-Kshot and 10way-Kshot configurations. In particular, STTMC outperforms other existing AMC methods by a large margin.
{"title":"STTMC: A Few-Shot Spatial Temporal Transductive Modulation Classifier","authors":"Yunhao Shi;Hua Xu;Zisen Qi;Yue Zhang;Dan Wang;Lei Jiang","doi":"10.1109/TMLCN.2024.3387430","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3387430","url":null,"abstract":"The advancement of deep learning (DL) techniques has led to significant progress in Automatic Modulation Classification (AMC). However, most existing DL-based AMC methods require massive training samples, which are difficult to obtain in non-cooperative scenarios. The identification of modulation types under small sample conditions has become an increasingly urgent problem. In this paper, we present a novel few-shot AMC model named the Spatial Temporal Transductive Modulation Classifier (STTMC), which comprises two modules: a feature extraction module and a graph network module. The former is responsible for extracting diverse features through a spatiotemporal parallel network, while the latter facilitates transductive decision-making through a graph network that uses a closed-form solution. Notably, STTMC classifies a group of test signals simultaneously to increase stability of few-shot model with an episode training strategy. Experimental results on the RadioML.2018.01A and RadioML.2016.10A datasets demonstrate that the proposed method perform well in 3way-Kshot, 5way-Kshot and 10way-Kshot configurations. In particular, STTMC outperforms other existing AMC methods by a large margin.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"546-559"},"PeriodicalIF":0.0,"publicationDate":"2024-04-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10497130","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140648000","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}
Pub Date : 2024-04-10DOI: 10.1109/TMLCN.2024.3386649
Asif Iqbal;Muhammad Naveed Aman;Biplab Sikdar
The Global Navigation Satellite System (GNSS) plays a crucial role in critical infrastructure by delivering precise timing and positional data. Nonetheless, the civilian segment of the GNSS remains susceptible to various spoofing attacks, necessitating robust detection mechanisms. The ability to deter such attacks significantly enhances the reliability and security of systems utilizing GNSS technology. Supervised Machine Learning (ML) techniques have shown promise in spoof detection. However, their effectiveness hinges on training data encompassing all possible attack scenarios, rendering them vulnerable to novel attack vectors. To address this limitation, we explore representation learning-based methods. These methods can be trained with a single data class and subsequently applied to classify test samples as either belonging to the training class or not. In this context, we introduce a GNSS spoof detection model comprising a Variational AutoEncoder (VAE) and a Generative Adversarial Network (GAN). The composite model is designed to efficiently learn the class distribution of the training data. The features used for training are extracted from the radio frequency and tracking modules of a standard GNSS receiver. To train our model, we leverage the Texas Spoofing Test Battery (TEXBAT) datasets. Our trained model yields three distinct detectors capable of effectively identifying spoofed signals. The detection performance across simpler to intermediate datasets for these detectors reaches approximately 99%, demonstrating their robustness. In the case of subtle attack scenario represented by DS-7, our approach achieves an approximate detection rate of 95%. In contrast, under supervised learning, the best detection score for DS-7 remains limited to 44.1%.
{"title":"A Deep Learning Based Induced GNSS Spoof Detection Framework","authors":"Asif Iqbal;Muhammad Naveed Aman;Biplab Sikdar","doi":"10.1109/TMLCN.2024.3386649","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3386649","url":null,"abstract":"The Global Navigation Satellite System (GNSS) plays a crucial role in critical infrastructure by delivering precise timing and positional data. Nonetheless, the civilian segment of the GNSS remains susceptible to various spoofing attacks, necessitating robust detection mechanisms. The ability to deter such attacks significantly enhances the reliability and security of systems utilizing GNSS technology. Supervised Machine Learning (ML) techniques have shown promise in spoof detection. However, their effectiveness hinges on training data encompassing all possible attack scenarios, rendering them vulnerable to novel attack vectors. To address this limitation, we explore representation learning-based methods. These methods can be trained with a single data class and subsequently applied to classify test samples as either belonging to the training class or not. In this context, we introduce a GNSS spoof detection model comprising a Variational AutoEncoder (VAE) and a Generative Adversarial Network (GAN). The composite model is designed to efficiently learn the class distribution of the training data. The features used for training are extracted from the radio frequency and tracking modules of a standard GNSS receiver. To train our model, we leverage the Texas Spoofing Test Battery (TEXBAT) datasets. Our trained model yields three distinct detectors capable of effectively identifying spoofed signals. The detection performance across simpler to intermediate datasets for these detectors reaches approximately 99%, demonstrating their robustness. In the case of subtle attack scenario represented by DS-7, our approach achieves an approximate detection rate of 95%. In contrast, under supervised learning, the best detection score for DS-7 remains limited to 44.1%.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"457-478"},"PeriodicalIF":0.0,"publicationDate":"2024-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10495074","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140633565","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}
In the past few years, Deep Reinforcement Learning (DRL) has become a valuable solution to automatically learn efficient resource management strategies in complex networks with time-varying statistics. However, the increased complexity of 5G and Beyond networks requires correspondingly more complex learning agents and the learning process itself might end up competing with users for communication and computational resources. This creates friction: on the one hand, the learning process needs resources to quickly converge to an effective strategy; on the other hand, the learning process needs to be efficient, i.e., take as few resources as possible from the user’s data plane, so as not to throttle users’ Quality of Service (QoS). In this paper, we investigate this trade-off, which we refer to as cost of learning, and propose a dynamic strategy to balance the resources assigned to the data plane and those reserved for learning. With the proposed approach, a learning agent can quickly converge to an efficient resource allocation strategy and adapt to changes in the environment as for the Continual Learning (CL) paradigm, while minimizing the impact on the users’ QoS. Simulation results show that the proposed method outperforms static allocation methods with minimal learning overhead, almost reaching the performance of an ideal out-of-band CL solution.
{"title":"Fast Context Adaptation in Cost-Aware Continual Learning","authors":"Seyyidahmed Lahmer;Federico Mason;Federico Chiariotti;Andrea Zanella","doi":"10.1109/TMLCN.2024.3386647","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3386647","url":null,"abstract":"In the past few years, Deep Reinforcement Learning (DRL) has become a valuable solution to automatically learn efficient resource management strategies in complex networks with time-varying statistics. However, the increased complexity of 5G and Beyond networks requires correspondingly more complex learning agents and the learning process itself might end up competing with users for communication and computational resources. This creates friction: on the one hand, the learning process needs resources to quickly converge to an effective strategy; on the other hand, the learning process needs to be efficient, i.e., take as few resources as possible from the user’s data plane, so as not to throttle users’ Quality of Service (QoS). In this paper, we investigate this trade-off, which we refer to as cost of learning, and propose a dynamic strategy to balance the resources assigned to the data plane and those reserved for learning. With the proposed approach, a learning agent can quickly converge to an efficient resource allocation strategy and adapt to changes in the environment as for the Continual Learning (CL) paradigm, while minimizing the impact on the users’ QoS. Simulation results show that the proposed method outperforms static allocation methods with minimal learning overhead, almost reaching the performance of an ideal out-of-band CL solution.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"479-494"},"PeriodicalIF":0.0,"publicationDate":"2024-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10495063","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140633566","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}
Pub Date : 2024-04-08DOI: 10.1109/TMLCN.2024.3385748
Abdulhamed Waraiet;Kanapathippillai Cumanan;Zhiguo Ding;Octavia A. Dobre
In this paper, we propose a robust design for an intelligent reflecting surface (IRS)-assisted multiple-input single output non-orthogonal multiple access (NOMA) system. By considering channel uncertainties, the original robust design problem is formulated as a sum-rate maximization problem under a set of constraints. In particular, the uncertainties associated with reflected channels through IRS elements and direct channels are taken into account in the design and they are modelled as bounded errors. However, the original robust problem is not jointly convex in terms of beamformers at the base station and phase shifts of IRS elements. Therefore, we reformulate the original robust design as a reinforcement learning problem and develop an algorithm based on the twin-delayed deep deterministic policy gradient agent (also known as TD3). In particular, the proposed algorithm solves the original problem by jointly designing the beamformers and the phase shifts, which is not possible with conventional optimization techniques. Numerical results are provided to validate the effectiveness and evaluate the performance of the proposed robust design. In particular, the results demonstrate the competitive and promising capabilities of the proposed robust algorithm, which achieves significant gains in terms of robustness and system sum-rates over the baseline deep deterministic policy gradient agent. In addition, the algorithm has the ability to deal with fixed and dynamic channels, which gives deep reinforcement learning methods an edge over hand-crafted convex optimization-based algorithms.
{"title":"Deep Reinforcement Learning-Based Robust Design for an IRS-Assisted MISO-NOMA System","authors":"Abdulhamed Waraiet;Kanapathippillai Cumanan;Zhiguo Ding;Octavia A. Dobre","doi":"10.1109/TMLCN.2024.3385748","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3385748","url":null,"abstract":"In this paper, we propose a robust design for an intelligent reflecting surface (IRS)-assisted multiple-input single output non-orthogonal multiple access (NOMA) system. By considering channel uncertainties, the original robust design problem is formulated as a sum-rate maximization problem under a set of constraints. In particular, the uncertainties associated with reflected channels through IRS elements and direct channels are taken into account in the design and they are modelled as bounded errors. However, the original robust problem is not jointly convex in terms of beamformers at the base station and phase shifts of IRS elements. Therefore, we reformulate the original robust design as a reinforcement learning problem and develop an algorithm based on the twin-delayed deep deterministic policy gradient agent (also known as TD3). In particular, the proposed algorithm solves the original problem by jointly designing the beamformers and the phase shifts, which is not possible with conventional optimization techniques. Numerical results are provided to validate the effectiveness and evaluate the performance of the proposed robust design. In particular, the results demonstrate the competitive and promising capabilities of the proposed robust algorithm, which achieves significant gains in terms of robustness and system sum-rates over the baseline deep deterministic policy gradient agent. In addition, the algorithm has the ability to deal with fixed and dynamic channels, which gives deep reinforcement learning methods an edge over hand-crafted convex optimization-based algorithms.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"424-441"},"PeriodicalIF":0.0,"publicationDate":"2024-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10494408","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140633557","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}
Pub Date : 2024-04-08DOI: 10.1109/TMLCN.2024.3386152
Swarna B. Chetty;Hamed Ahmadi;Avishek Nag
The sixth generation of mobile networks (6G) promises applications and services with faster data rates, ultra-reliability, and lower latency compared to the fifth-generation mobile networks (5G). These highly demanding 6G applications will burden the network by imposing stringent performance requirements. Network Function Virtualization (NFV) reduces costs by running network functions as Virtual Network Functions (VNFs) on commodity hardware. While NFV is a promising solution, it poses Resource Allocation (RA) challenges. To enhance RA efficiency, we addressed two critical subproblems: the requirement of dynamic service priority and a low-priority service starvation problem. We introduce ‘Dynamic Prioritization’ (DyPr), employing an ML model to emphasize macro- and microlevel priority for unseen services and address the existing starvation problem in current solutions and their limitations. We present ‘Adaptive Scheduling’ (AdSch), a three-factor approach (priority, threshold waiting time, and reliability) that surpasses traditional priority-based methods. In this context, starvation refers to extended waiting times and the eventual rejection of low-priority services due to a ‘delay. Also, to further investigate, a traffic-aware starvation and deployment problem is studied to enhance efficiency. We employed a Deep Deterministic Policy Gradient (DDPG) model for adaptive scheduling and an online Ridge Regression (RR) model for dynamic prioritization, creating a zero-touch solution. The DDPG model efficiently identified ‘Beneficial and Starving’ services, alleviating the starvation issue by deploying twice as many low-priority services. With an accuracy rate exceeding 80%, our online RR model quickly learns prioritization patterns in under 100 transitions. We categorized services as ‘High-Demand’ (HD) or ‘Not So High Demand’ (NHD) based on traffic volume, providing insight into high revenue-generating services. We achieved a nearly optimal resource allocation by balancing low-priority HD and low-priority NHD services, deploying twice as many low-priority HD services as a model without traffic awareness.
与第五代移动网络（5G）相比，第六代移动网络（6G）承诺提供更快的数据传输速率、超高的可靠性和更低的延迟的应用和服务。这些高要求的 6G 应用将对网络提出严格的性能要求，从而加重网络负担。网络功能虚拟化（NFV）通过在商品硬件上以虚拟网络功能（VNF）的形式运行网络功能来降低成本。虽然 NFV 是一种前景广阔的解决方案，但它也带来了资源分配（RA）方面的挑战。为了提高资源分配效率，我们解决了两个关键的子问题：动态服务优先级要求和低优先级服务饥饿问题。我们引入了 "动态优先级"（DyPr），采用 ML 模型来强调未见服务的宏观和微观优先级，并解决目前解决方案中存在的饥饿问题及其局限性。我们提出了 "自适应调度"（AdSch），这是一种三因素方法（优先级、阈值等待时间和可靠性），超越了传统的基于优先级的方法。在这里，"饥饿 "指的是等待时间延长，以及由于 "延迟 "而最终拒绝低优先级服务。此外，为了进一步研究，我们还研究了流量感知的饥饿和部署问题，以提高效率。我们采用了用于自适应调度的深度确定性策略梯度（DDPG）模型和用于动态优先级排序的在线岭回归（RR）模型，创建了一个零接触解决方案。DDPG 模型能有效识别 "受益和饥饿 "服务，通过部署两倍的低优先级服务来缓解饥饿问题。我们的在线 RR 模型准确率超过 80%，可在 100 次转换中快速学习优先级模式。我们根据流量将服务分为 "高需求"（HD）和 "非高需求"（NHD），以便深入了解高创收服务。通过平衡低优先级 HD 服务和低优先级 NHD 服务，我们实现了近乎最优的资源分配，部署的低优先级 HD 服务数量是无流量感知模型的两倍。
{"title":"A DDPG-Based Zero-Touch Dynamic Prioritization to Address Starvation of Services for Deploying Microservices-Based VNFs","authors":"Swarna B. Chetty;Hamed Ahmadi;Avishek Nag","doi":"10.1109/TMLCN.2024.3386152","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3386152","url":null,"abstract":"The sixth generation of mobile networks (6G) promises applications and services with faster data rates, ultra-reliability, and lower latency compared to the fifth-generation mobile networks (5G). These highly demanding 6G applications will burden the network by imposing stringent performance requirements. Network Function Virtualization (NFV) reduces costs by running network functions as Virtual Network Functions (VNFs) on commodity hardware. While NFV is a promising solution, it poses Resource Allocation (RA) challenges. To enhance RA efficiency, we addressed two critical subproblems: the requirement of dynamic service priority and a low-priority service starvation problem. We introduce ‘Dynamic Prioritization’ (DyPr), employing an ML model to emphasize macro- and microlevel priority for unseen services and address the existing starvation problem in current solutions and their limitations. We present ‘Adaptive Scheduling’ (AdSch), a three-factor approach (priority, threshold waiting time, and reliability) that surpasses traditional priority-based methods. In this context, starvation refers to extended waiting times and the eventual rejection of low-priority services due to a ‘delay. Also, to further investigate, a traffic-aware starvation and deployment problem is studied to enhance efficiency. We employed a Deep Deterministic Policy Gradient (DDPG) model for adaptive scheduling and an online Ridge Regression (RR) model for dynamic prioritization, creating a zero-touch solution. The DDPG model efficiently identified ‘Beneficial and Starving’ services, alleviating the starvation issue by deploying twice as many low-priority services. With an accuracy rate exceeding 80%, our online RR model quickly learns prioritization patterns in under 100 transitions. We categorized services as ‘High-Demand’ (HD) or ‘Not So High Demand’ (NHD) based on traffic volume, providing insight into high revenue-generating services. We achieved a nearly optimal resource allocation by balancing low-priority HD and low-priority NHD services, deploying twice as many low-priority HD services as a model without traffic awareness.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"526-545"},"PeriodicalIF":0.0,"publicationDate":"2024-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10494765","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140647819","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}
Pub Date : 2024-04-04DOI: 10.1109/TMLCN.2024.3385355
Jie Zhang;Li Chen;Yunfei Chen;Xiaohui Chen;Guo Wei
Decentralized federated learning (DFL) architecture enables clients to collaboratively train a shared machine learning model without a central parameter server. However, it is difficult to apply DFL to a multi-cell scenario due to inadequate model averaging and cross-cell device-to-device (D2D) communications. In this paper, we propose a hierarchically decentralized federated learning (HDFL) framework that combines intra-cell D2D links between devices and backhaul communications between base stations. In HDFL, devices from different cells collaboratively train a global model using periodic intra-cell D2D consensus and inter-cell aggregation. The strong convergence guarantee of the proposed HDFL algorithm is established even for non-convex objectives. Based on the convergence analysis, we characterize the network topology of each cell, the communication interval of intra-cell consensus and inter-cell aggregation on the training performance. To further improve the performance of HDFL, we optimize the computation capacity selection and bandwidth allocation to minimize the training latency and energy overhead. Numerical results based on the MNIST and CIFAR-10 datasets validate the superiority of HDFL over traditional DFL methods in the multi-cell scenario.
{"title":"Hierarchically Federated Learning in Wireless Networks: D2D Consensus and Inter-Cell Aggregation","authors":"Jie Zhang;Li Chen;Yunfei Chen;Xiaohui Chen;Guo Wei","doi":"10.1109/TMLCN.2024.3385355","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3385355","url":null,"abstract":"Decentralized federated learning (DFL) architecture enables clients to collaboratively train a shared machine learning model without a central parameter server. However, it is difficult to apply DFL to a multi-cell scenario due to inadequate model averaging and cross-cell device-to-device (D2D) communications. In this paper, we propose a hierarchically decentralized federated learning (HDFL) framework that combines intra-cell D2D links between devices and backhaul communications between base stations. In HDFL, devices from different cells collaboratively train a global model using periodic intra-cell D2D consensus and inter-cell aggregation. The strong convergence guarantee of the proposed HDFL algorithm is established even for non-convex objectives. Based on the convergence analysis, we characterize the network topology of each cell, the communication interval of intra-cell consensus and inter-cell aggregation on the training performance. To further improve the performance of HDFL, we optimize the computation capacity selection and bandwidth allocation to minimize the training latency and energy overhead. Numerical results based on the MNIST and CIFAR-10 datasets validate the superiority of HDFL over traditional DFL methods in the multi-cell scenario.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"442-456"},"PeriodicalIF":0.0,"publicationDate":"2024-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10491307","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140633558","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}
Pub Date : 2024-04-02DOI: 10.1109/TMLCN.2024.3384329
Wei Cui;Wei Yu
In most applications of utilizing neural networks for mathematical optimization, a dedicated model is trained for each specific optimization objective. However, in many scenarios, several distinct yet correlated objectives or tasks often need to be optimized on the same set of problem inputs. Instead of independently training a different neural network for each problem separately, it would be more efficient to exploit the correlations between these objectives and to train multiple neural network models with shared model parameters and feature representations. To achieve this, this paper first establishes the concept of common information: the shared knowledge required for solving the correlated tasks, then proposes a novel approach for model training by adding into the model an additional reconstruction stage associated with a new reconstruction loss. This loss is for reconstructing the common information starting from a selected hidden layer in the model. The proposed approach encourages the learned features to be general and transferable, and therefore can be readily used for efficient transfer learning. For numerical simulations, three applications are studied: transfer learning on classifying MNIST handwritten digits, the device-to-device wireless network power allocation, and the multiple-input-single-output network downlink beamforming and localization. Simulation results suggest that the proposed approach is highly efficient in data and model complexity, is resilient to over-fitting, and has competitive performances.
{"title":"Transfer Learning With Reconstruction Loss","authors":"Wei Cui;Wei Yu","doi":"10.1109/TMLCN.2024.3384329","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3384329","url":null,"abstract":"In most applications of utilizing neural networks for mathematical optimization, a dedicated model is trained for each specific optimization objective. However, in many scenarios, several distinct yet correlated objectives or tasks often need to be optimized on the same set of problem inputs. Instead of independently training a different neural network for each problem separately, it would be more efficient to exploit the correlations between these objectives and to train multiple neural network models with shared model parameters and feature representations. To achieve this, this paper first establishes the concept of common information: the shared knowledge required for solving the correlated tasks, then proposes a novel approach for model training by adding into the model an additional reconstruction stage associated with a new reconstruction loss. This loss is for reconstructing the common information starting from a selected hidden layer in the model. The proposed approach encourages the learned features to be general and transferable, and therefore can be readily used for efficient transfer learning. For numerical simulations, three applications are studied: transfer learning on classifying MNIST handwritten digits, the device-to-device wireless network power allocation, and the multiple-input-single-output network downlink beamforming and localization. Simulation results suggest that the proposed approach is highly efficient in data and model complexity, is resilient to over-fitting, and has competitive performances.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"407-423"},"PeriodicalIF":0.0,"publicationDate":"2024-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10488445","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140633559","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}
Pub Date : 2024-03-31DOI: 10.1109/TMLCN.2024.3408131
Feng Wang;M. Cenk Gursoy;Senem Velipasalar
In this paper, we propose feature-based federated transfer learning as a novel approach to improve communication efficiency by reducing the uplink payload by multiple orders of magnitude compared to that of existing approaches in federated learning and federated transfer learning. Specifically, in the proposed feature-based federated learning, we design the extracted features and outputs to be uploaded instead of parameter updates. For this distributed learning model, we determine the required payload and provide comparisons with the existing schemes. Subsequently, we analyze the robustness of feature-based federated transfer learning against packet loss, data insufficiency, and quantization. Finally, we address privacy considerations by defining and analyzing label privacy leakage and feature privacy leakage, and investigating mitigating approaches. For all aforementioned analyses, we evaluate the performance of the proposed learning scheme via experiments on an image classification task and a natural language processing task to demonstrate its effectiveness ( https://github.com/wfwf10/Feature-based-Federated-Transfer-Learning ).
{"title":"Feature-Based Federated Transfer Learning: Communication Efficiency, Robustness and Privacy","authors":"Feng Wang;M. Cenk Gursoy;Senem Velipasalar","doi":"10.1109/TMLCN.2024.3408131","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3408131","url":null,"abstract":"In this paper, we propose feature-based federated transfer learning as a novel approach to improve communication efficiency by reducing the uplink payload by multiple orders of magnitude compared to that of existing approaches in federated learning and federated transfer learning. Specifically, in the proposed feature-based federated learning, we design the extracted features and outputs to be uploaded instead of parameter updates. For this distributed learning model, we determine the required payload and provide comparisons with the existing schemes. Subsequently, we analyze the robustness of feature-based federated transfer learning against packet loss, data insufficiency, and quantization. Finally, we address privacy considerations by defining and analyzing label privacy leakage and feature privacy leakage, and investigating mitigating approaches. For all aforementioned analyses, we evaluate the performance of the proposed learning scheme via experiments on an image classification task and a natural language processing task to demonstrate its effectiveness (\u0000<uri>https://github.com/wfwf10/Feature-based-Federated-Transfer-Learning</uri>\u0000).","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"823-840"},"PeriodicalIF":0.0,"publicationDate":"2024-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10542971","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141453361","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}
Pub Date : 2024-03-21DOI: 10.1109/TMLCN.2024.3403789
Pialy Biswas;Ranjan K. Mallik;Khaled B. Letaief
This paper proposes a Gaussian mixture model (GMM) based access point (AP) clustering technique in cell-free massive MIMO (CFMM) communication systems. The APs are first clustered on the basis of large-scale fading coefficients, and the users are assigned to each cluster depending on the channel gain. As the number of clusters increases, there is a degradation in the overall data rate of the system, causing a trade-off between the cluster number and average rate per user. To address this problem, we present an optimization problem that optimizes both the upper bound on the average downlink rate per user and the number of clusters. The optimal number of clusters is intuitively determined by solving the optimization problem, and then grouping the APs and users. As a result, the computation expense is much lower than the current techniques, since the existing methods require evaluations of the network performance in multiple iterations to find the optimal number of clusters. In addition, we analyze the performance of both balanced and unbalanced clustering. Numerical results will indicate that the unbalanced clustering yields a superior rate per user while maintaining a lower level of complexity compared to the balanced one. Furthermore, we investigate the statistical analysis of the spectral efficiency (SE) per user in the clustered CFMM. The findings reveal that the SE per user can be approximated by the logistic distribution.
{"title":"Optimal Access Point Centric Clustering for Cell-Free Massive MIMO Using Gaussian Mixture Model Clustering","authors":"Pialy Biswas;Ranjan K. Mallik;Khaled B. Letaief","doi":"10.1109/TMLCN.2024.3403789","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3403789","url":null,"abstract":"This paper proposes a Gaussian mixture model (GMM) based access point (AP) clustering technique in cell-free massive MIMO (CFMM) communication systems. The APs are first clustered on the basis of large-scale fading coefficients, and the users are assigned to each cluster depending on the channel gain. As the number of clusters increases, there is a degradation in the overall data rate of the system, causing a trade-off between the cluster number and average rate per user. To address this problem, we present an optimization problem that optimizes both the upper bound on the average downlink rate per user and the number of clusters. The optimal number of clusters is intuitively determined by solving the optimization problem, and then grouping the APs and users. As a result, the computation expense is much lower than the current techniques, since the existing methods require evaluations of the network performance in multiple iterations to find the optimal number of clusters. In addition, we analyze the performance of both balanced and unbalanced clustering. Numerical results will indicate that the unbalanced clustering yields a superior rate per user while maintaining a lower level of complexity compared to the balanced one. Furthermore, we investigate the statistical analysis of the spectral efficiency (SE) per user in the clustered CFMM. The findings reveal that the SE per user can be approximated by the logistic distribution.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"675-687"},"PeriodicalIF":0.0,"publicationDate":"2024-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10535986","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141187291","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：