Semantic and goal-oriented (SGO) communication is an emerging technology that only transmits significant information for a given task. Semantic communication encounters many challenges, such as computational complexity at end users, availability of data, and privacy-preserving. This work presents a TinyML-based semantic communication framework for few-shot wireless image classification that integrates split-learning and meta-learning. We exploit split-learning to limit the computations performed by the end-users while ensuring privacy-preserving. In addition, meta-learning overcomes data availability concerns and speeds up training by utilizing similarly trained tasks. The proposed algorithm is tested using a data set of images of hand-written letters. In addition, we present an uncertainty analysis of the predictions using conformal prediction (CP) techniques. Simulation results show that the proposed Semantic-MSL outperforms conventional schemes by achieving a ${20} %$ gain in classification accuracy using fewer data points yet less training energy consumption.
{"title":"Semantic Meta-Split Learning: A TinyML Scheme for Few-Shot Wireless Image Classification","authors":"Eslam Eldeeb;Mohammad Shehab;Hirley Alves;Mohamed-Slim Alouini","doi":"10.1109/TMLCN.2025.3557734","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3557734","url":null,"abstract":"Semantic and goal-oriented (SGO) communication is an emerging technology that only transmits significant information for a given task. Semantic communication encounters many challenges, such as computational complexity at end users, availability of data, and privacy-preserving. This work presents a TinyML-based semantic communication framework for few-shot wireless image classification that integrates split-learning and meta-learning. We exploit split-learning to limit the computations performed by the end-users while ensuring privacy-preserving. In addition, meta-learning overcomes data availability concerns and speeds up training by utilizing similarly trained tasks. The proposed algorithm is tested using a data set of images of hand-written letters. In addition, we present an uncertainty analysis of the predictions using conformal prediction (CP) techniques. Simulation results show that the proposed Semantic-MSL outperforms conventional schemes by achieving a <inline-formula> <tex-math>${20} %$ </tex-math></inline-formula> gain in classification accuracy using fewer data points yet less training energy consumption.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"491-501"},"PeriodicalIF":0.0,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10948463","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817929","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 : 2025-03-31DOI: 10.1109/TMLCN.2025.3555975
Ke He;Thang Xuan Vu;Lisheng Fan;Symeon Chatzinotas;Björn Ottersten
This paper investigates the spatio-temporal predictive learning problem, which is crucial in diverse applications such as MIMO channel prediction, mobile traffic analysis, and network slicing. To address this problem, the attention mechanism has been adopted by many existing models to predict future outputs. However, most of these models use a single-domain attention which captures input dependency structures only in the temporal domain. This limitation reduces their prediction accuracy in spatio-temporal predictive learning, where understanding both spatial and temporal dependencies is essential. To tackle this issue and enhance the prediction performance, we propose a novel crossover attention mechanism in this paper. The crossover attention can be understood as a learnable regression kernel which prioritizes the input sequence with both spatial and temporal similarities and extracts relevant information for generating the output of future time slots. Simulation results and ablation studies based on synthetic and realistic datasets show that the proposed crossover attention achieves considerable prediction accuracy improvement compared to the conventional attention layers.
{"title":"Spatio-Temporal Predictive Learning Using Crossover Attention for Communications and Networking Applications","authors":"Ke He;Thang Xuan Vu;Lisheng Fan;Symeon Chatzinotas;Björn Ottersten","doi":"10.1109/TMLCN.2025.3555975","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3555975","url":null,"abstract":"This paper investigates the spatio-temporal predictive learning problem, which is crucial in diverse applications such as MIMO channel prediction, mobile traffic analysis, and network slicing. To address this problem, the attention mechanism has been adopted by many existing models to predict future outputs. However, most of these models use a single-domain attention which captures input dependency structures only in the temporal domain. This limitation reduces their prediction accuracy in spatio-temporal predictive learning, where understanding both spatial and temporal dependencies is essential. To tackle this issue and enhance the prediction performance, we propose a novel crossover attention mechanism in this paper. The crossover attention can be understood as a learnable regression kernel which prioritizes the input sequence with both spatial and temporal similarities and extracts relevant information for generating the output of future time slots. Simulation results and ablation studies based on synthetic and realistic datasets show that the proposed crossover attention achieves considerable prediction accuracy improvement compared to the conventional attention layers.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"479-490"},"PeriodicalIF":0.0,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10945971","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817995","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 : 2025-03-31DOI: 10.1109/TMLCN.2025.3556634
Dengyu Wu;Jiechen Chen;Bipin Rajendran;H. Vincent Poor;Osvaldo Simeone
Inspired by biological processes, neuromorphic computing leverages spiking neural networks (SNNs) to perform inference tasks, offering significant efficiency gains for workloads involving sequential data. Recent advances in hardware and software have shown that embedding a small payload within each spike exchanged between spiking neurons can enhance inference accuracy without increasing energy consumption. To scale neuromorphic computing to larger workloads, split computing—where an SNN is partitioned across two devices—is a promising solution. In such architectures, the device hosting the initial layers must transmit information about the spikes generated by its output neurons to the second device. This establishes a trade-off between the benefits of multi-level spikes, which carry additional payload information, and the communication resources required for transmitting extra bits between devices. This paper presents the first comprehensive study of a neuromorphic wireless split computing architecture that employs multi-level SNNs. We propose digital and analog modulation schemes for an orthogonal frequency division multiplexing (OFDM) radio interface to enable efficient communication. Simulation and experimental results using software-defined radios reveal performance improvements achieved by multi-level SNN models and provide insights into the optimal payload size as a function of the connection quality between the transmitter and receiver.
{"title":"Neuromorphic Wireless Split Computing With Multi-Level Spikes","authors":"Dengyu Wu;Jiechen Chen;Bipin Rajendran;H. Vincent Poor;Osvaldo Simeone","doi":"10.1109/TMLCN.2025.3556634","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3556634","url":null,"abstract":"Inspired by biological processes, neuromorphic computing leverages spiking neural networks (SNNs) to perform inference tasks, offering significant efficiency gains for workloads involving sequential data. Recent advances in hardware and software have shown that embedding a small payload within each spike exchanged between spiking neurons can enhance inference accuracy without increasing energy consumption. To scale neuromorphic computing to larger workloads, split computing—where an SNN is partitioned across two devices—is a promising solution. In such architectures, the device hosting the initial layers must transmit information about the spikes generated by its output neurons to the second device. This establishes a trade-off between the benefits of multi-level spikes, which carry additional payload information, and the communication resources required for transmitting extra bits between devices. This paper presents the first comprehensive study of a neuromorphic wireless split computing architecture that employs multi-level SNNs. We propose digital and analog modulation schemes for an orthogonal frequency division multiplexing (OFDM) radio interface to enable efficient communication. Simulation and experimental results using software-defined radios reveal performance improvements achieved by multi-level SNN models and provide insights into the optimal payload size as a function of the connection quality between the transmitter and receiver.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"502-516"},"PeriodicalIF":0.0,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10946192","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143835442","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 : 2025-03-30DOI: 10.1109/TMLCN.2025.3575368
Kazi Hasan;Khaleda Papry;Thomas Trappenberg;Israat Haque
Radio Link Failure (RLF) prediction system in Radio Access Networks (RANs) is critical for ensuring seamless communication and meeting the stringent requirements of high data rates, low latency, and improved reliability in 5G networks. However, weather conditions such as precipitation, humidity, temperature, and wind impact these communication links. Usually, historical radio link Key Performance Indicators (KPIs) and their surrounding weather station observations are utilized for building learning-based RLF prediction models. However, such models must be capable of learning the spatial weather context in a dynamic RAN and effectively encoding time series KPIs with the weather observation data. Existing work utilizes a heuristic-based and non-generalizable weather station aggregation method that uses Long Short-Term Memory (LSTM) for non-weighted sequence modeling. This paper fills the gap by proposing GenTrap, a novel RLF prediction framework that introduces a Graph Neural Network (GNN)-based learnable weather effect aggregation module and employs state-of-the-art time series transformer as the temporal feature extractor for radio link failure prediction. The GNN module encodes surrounding weather station data of each radio site while the transformer module encodes historical radio and weather observation features. The proposed aggregation method of GenTrap can be integrated into any existing prediction model to achieve better performance and generalizability. We evaluate GenTrap on two real-world datasets (rural and urban) with 2.6 million KPI data points and show that GenTrap offers a significantly higher F1-score of 0.93 for rural and 0.79 for urban, an increase of 29% and 21% respectively, compared to the state-of-the-art LSTM-based solutions while offering a 20% increased generalization capability.
无线接入网（RANs）中的RLF （Radio Link Failure）预测系统对于确保无缝通信，满足5G网络对高数据速率、低延迟和高可靠性的严格要求至关重要。然而，诸如降水、湿度、温度和风等天气条件会影响这些通信链路。通常，利用历史无线电链路关键性能指标（kpi）及其周围气象站观测数据建立基于学习的RLF预测模型。然而，这种模型必须能够在动态RAN中学习空间天气环境，并有效地用天气观测数据编码时间序列kpi。现有工作采用基于启发式的非一般化气象站聚合方法，该方法使用长短期记忆（LSTM）进行非加权序列建模。本文提出了一种新的RLF预测框架GenTrap，该框架引入了基于图神经网络（GNN）的可学习天气效应聚合模块，并采用最先进的时间序列变压器作为无线电链路故障预测的时间特征提取器，填补了这一空白。GNN模块对每个无线电站点周围气象站数据进行编码，变压器模块对历史无线电和天气观测特征进行编码。所提出的GenTrap聚合方法可以集成到任何现有的预测模型中，以获得更好的性能和泛化性。我们在两个真实世界的数据集（农村和城市）上使用260万KPI数据点对GenTrap进行了评估，结果表明，与最先进的基于lstm的解决方案相比，GenTrap在农村和城市的f1得分分别为0.93和0.79，分别提高了29%和21%，同时泛化能力提高了20%。
{"title":"A Generalized GNN-Transformer-Based Radio Link Failure Prediction Framework in 5G RAN","authors":"Kazi Hasan;Khaleda Papry;Thomas Trappenberg;Israat Haque","doi":"10.1109/TMLCN.2025.3575368","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3575368","url":null,"abstract":"Radio Link Failure (RLF) prediction system in Radio Access Networks (RANs) is critical for ensuring seamless communication and meeting the stringent requirements of high data rates, low latency, and improved reliability in 5G networks. However, weather conditions such as precipitation, humidity, temperature, and wind impact these communication links. Usually, historical radio link Key Performance Indicators (KPIs) and their surrounding weather station observations are utilized for building learning-based RLF prediction models. However, such models must be capable of learning the spatial weather context in a dynamic RAN and effectively encoding time series KPIs with the weather observation data. Existing work utilizes a heuristic-based and non-generalizable weather station aggregation method that uses Long Short-Term Memory (LSTM) for non-weighted sequence modeling. This paper fills the gap by proposing GenTrap, a novel RLF prediction framework that introduces a Graph Neural Network (GNN)-based learnable weather effect aggregation module and employs state-of-the-art time series transformer as the temporal feature extractor for radio link failure prediction. The GNN module encodes surrounding weather station data of each radio site while the transformer module encodes historical radio and weather observation features. The proposed aggregation method of GenTrap can be integrated into any existing prediction model to achieve better performance and generalizability. We evaluate GenTrap on two real-world datasets (rural and urban) with 2.6 million KPI data points and show that GenTrap offers a significantly higher F1-score of 0.93 for rural and 0.79 for urban, an increase of 29% and 21% respectively, compared to the state-of-the-art LSTM-based solutions while offering a 20% increased generalization capability.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"710-724"},"PeriodicalIF":0.0,"publicationDate":"2025-03-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11018489","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144308387","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 : 2025-03-20DOI: 10.1109/TMLCN.2025.3553100
Berend J. D. Gort;Godfrey M. Kibalya;Angelos Antonopoulos
Effective resource management in edge-cloud networks is crucial for meeting Quality of Service (QoS) requirements while minimizing operational costs. However, dynamic and fluctuating workloads pose significant challenges for accurate workload prediction and efficient resource allocation, particularly in resource-constrained edge environments. In this paper, we introduce AERO (Adaptive Edge-cloud Resource Orchestration), a novel lightweight forecasting model designed to address these challenges. AERO features an adaptive period detection mechanism that dynamically identifies dominant periodicities in multivariate workload data, allowing it to adjust to varying patterns and abrupt changes. With fewer than 1,000 parameters, AERO is highly suitable for deployment on edge devices with limited computational capacity. We formalize our approach through a comprehensive system model and extend an existing simulation framework with predictor modules to evaluate AERO’s performance in realistic cloud-edge environments. Our extensive evaluations on real-world cloud workload datasets demonstrate that AERO achieves comparable prediction accuracy to complex state-of-the-art models with millions of parameters, while significantly reducing model size and computational overhead. In addition, simulations show that AERO improves orchestration performance, reducing energy consumption and response times compared to existing proactive and reactive approaches. Our live deployment experiments further validate these findings, demonstrating that AERO consistently delivers superior performance. These results highlight AERO as an effective solution for improving resource management and reducing operational costs in dynamic cloud-edge environments.
{"title":"AERO: Adaptive Edge-Cloud Orchestration With a Sub-1K-Parameter Forecasting Model","authors":"Berend J. D. Gort;Godfrey M. Kibalya;Angelos Antonopoulos","doi":"10.1109/TMLCN.2025.3553100","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3553100","url":null,"abstract":"Effective resource management in edge-cloud networks is crucial for meeting Quality of Service (QoS) requirements while minimizing operational costs. However, dynamic and fluctuating workloads pose significant challenges for accurate workload prediction and efficient resource allocation, particularly in resource-constrained edge environments. In this paper, we introduce AERO (Adaptive Edge-cloud Resource Orchestration), a novel lightweight forecasting model designed to address these challenges. AERO features an adaptive period detection mechanism that dynamically identifies dominant periodicities in multivariate workload data, allowing it to adjust to varying patterns and abrupt changes. With fewer than 1,000 parameters, AERO is highly suitable for deployment on edge devices with limited computational capacity. We formalize our approach through a comprehensive system model and extend an existing simulation framework with predictor modules to evaluate AERO’s performance in realistic cloud-edge environments. Our extensive evaluations on real-world cloud workload datasets demonstrate that AERO achieves comparable prediction accuracy to complex state-of-the-art models with millions of parameters, while significantly reducing model size and computational overhead. In addition, simulations show that AERO improves orchestration performance, reducing energy consumption and response times compared to existing proactive and reactive approaches. Our live deployment experiments further validate these findings, demonstrating that AERO consistently delivers superior performance. These results highlight AERO as an effective solution for improving resource management and reducing operational costs in dynamic cloud-edge environments.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"463-478"},"PeriodicalIF":0.0,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10935743","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143740375","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 : 2025-03-19DOI: 10.1109/TMLCN.2025.3571026
Michael Baur;Nurettin Turan;Simon Wallner;Wolfgang Utschick
Generative models are typically evaluated by direct inspection of their generated samples, e.g., by visual inspection in the case of images. Further evaluation metrics like the Fréchet inception distance or maximum mean discrepancy are intricate to interpret and lack physical motivation. These observations make evaluating generative models in the wireless PHY layer non-trivial. This work establishes a framework consisting of evaluation metrics and methods for generative models applied to the wireless PHY layer. The proposed metrics and methods are motivated by wireless applications, facilitating interpretation and understandability for the wireless community. In particular, we propose a spectral efficiency analysis for validating the generated channel norms and a codebook fingerprinting method to validate the generated channel directions. Moreover, we propose an application cross-check to evaluate the generative model’s samples for training machine learning-based models in relevant downstream tasks. Our analysis is based on real-world measurement data and includes the Gaussian mixture model, variational autoencoder, diffusion model, and generative adversarial network. Our results indicate that solely relying on metrics like the maximum mean discrepancy produces inconsistent and uninterpretable evaluation outcomes. In contrast, the proposed metrics and methods exhibit consistent and explainable behavior.
{"title":"Evaluation Metrics and Methods for Generative Models in the Wireless PHY Layer","authors":"Michael Baur;Nurettin Turan;Simon Wallner;Wolfgang Utschick","doi":"10.1109/TMLCN.2025.3571026","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3571026","url":null,"abstract":"Generative models are typically evaluated by direct inspection of their generated samples, e.g., by visual inspection in the case of images. Further evaluation metrics like the Fréchet inception distance or maximum mean discrepancy are intricate to interpret and lack physical motivation. These observations make evaluating generative models in the wireless PHY layer non-trivial. This work establishes a framework consisting of evaluation metrics and methods for generative models applied to the wireless PHY layer. The proposed metrics and methods are motivated by wireless applications, facilitating interpretation and understandability for the wireless community. In particular, we propose a spectral efficiency analysis for validating the generated channel norms and a codebook fingerprinting method to validate the generated channel directions. Moreover, we propose an application cross-check to evaluate the generative model’s samples for training machine learning-based models in relevant downstream tasks. Our analysis is based on real-world measurement data and includes the Gaussian mixture model, variational autoencoder, diffusion model, and generative adversarial network. Our results indicate that solely relying on metrics like the maximum mean discrepancy produces inconsistent and uninterpretable evaluation outcomes. In contrast, the proposed metrics and methods exhibit consistent and explainable behavior.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"677-689"},"PeriodicalIF":0.0,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11007069","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144219745","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 : 2025-03-18DOI: 10.1109/TMLCN.2025.3551689
Sepideh Afshar;Reza Razavi;Mohammad Moshirpour
Accurate throughput forecasting is essential for ensuring the seamless operation of Real-Time Communication (RTC) applications. These demands for accurate throughput forecasting become particularly challenging when dealing with wireless access links, as they inherently exhibit fluctuating bandwidth. Ensuring an exceptional user Quality of Experience (QoE) in this scenario depends on accurately predicting available bandwidth in the short term since it plays a pivotal role in guiding video rate adaptation. Yet, current methodologies for short-term bandwidth prediction (SBP) struggle to perform adequately in dynamically changing real-world network environments and lack generalizability to adapt across varied network conditions. Also, acquiring long and representative traces that capture real-world network complexity is challenging. To overcome these challenges, we propose closed-loop clustering-based Global Forecasting Models (GFMs) for SBP. Unlike local models, GFMs apply the same function to all traces enabling cross-learning, and leveraging relationships among traces to address the performance issues seen in current SBP algorithms. To address potential heterogeneity within the data and improve prediction quality, a clustered-wise GFM is utilized to group similar traces based on prediction accuracy. Finally, the proposed method is validated using real-world datasets of HSDPA 3G, NYC LTE, and Irish 5G data demonstrating significant improvements in accuracy and generalizability.
{"title":"Closed-Loop Clustering-Based Global Bandwidth Prediction in Real-Time Video Streaming","authors":"Sepideh Afshar;Reza Razavi;Mohammad Moshirpour","doi":"10.1109/TMLCN.2025.3551689","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3551689","url":null,"abstract":"Accurate throughput forecasting is essential for ensuring the seamless operation of Real-Time Communication (RTC) applications. These demands for accurate throughput forecasting become particularly challenging when dealing with wireless access links, as they inherently exhibit fluctuating bandwidth. Ensuring an exceptional user Quality of Experience (QoE) in this scenario depends on accurately predicting available bandwidth in the short term since it plays a pivotal role in guiding video rate adaptation. Yet, current methodologies for short-term bandwidth prediction (SBP) struggle to perform adequately in dynamically changing real-world network environments and lack generalizability to adapt across varied network conditions. Also, acquiring long and representative traces that capture real-world network complexity is challenging. To overcome these challenges, we propose closed-loop clustering-based Global Forecasting Models (GFMs) for SBP. Unlike local models, GFMs apply the same function to all traces enabling cross-learning, and leveraging relationships among traces to address the performance issues seen in current SBP algorithms. To address potential heterogeneity within the data and improve prediction quality, a clustered-wise GFM is utilized to group similar traces based on prediction accuracy. Finally, the proposed method is validated using real-world datasets of HSDPA 3G, NYC LTE, and Irish 5G data demonstrating significant improvements in accuracy and generalizability.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"448-462"},"PeriodicalIF":0.0,"publicationDate":"2025-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10929655","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143716486","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 : 2025-03-10DOI: 10.1109/TMLCN.2025.3550119
Ce Feng;Parv Venkitasubramaniam
The increasing adoption of machine learning at the edge (ML-at-the-edge) and federated learning (FL) presents a dual challenge: ensuring data privacy as well as addressing resource constraints such as limited computational power, memory, and communication bandwidth. Traditional approaches typically apply differentially private stochastic gradient descent (DP-SGD) to preserve privacy, followed by quantization techniques as a post-processing step to reduce model size and communication overhead. However, this sequential framework introduces inherent drawbacks, as quantization alone lacks privacy guarantees and often introduces errors that degrade model performance. In this work, we propose randomized quantization as an integrated solution to address these dual challenges by embedding randomness directly into the quantization process. This approach enhances privacy while simultaneously reducing communication and computational overhead. To achieve this, we introduce Randomized Quantizer Projection Stochastic Gradient Descent (RQP-SGD), a method designed for ML-at-the-edge that embeds DP-SGD within a randomized quantization-based projection during model training. For federated learning, we develop Gaussian Sampling Quantization (GSQ), which integrates discrete Gaussian sampling into the quantization process to ensure local differential privacy (LDP). Unlike conventional methods that rely on Gaussian noise addition, GSQ achieves privacy through discrete Gaussian sampling while improving communication efficiency and model utility across distributed systems. Through rigorous theoretical analysis and extensive experiments on benchmark datasets, we demonstrate that these methods significantly enhance the utility-privacy trade-off and computational efficiency in both ML-at-the-edge and FL systems. RQP-SGD is evaluated on MNIST and the Breast Cancer Diagnostic dataset, showing an average 10.62% utility improvement over the deterministic quantization-based projected DP-SGD while maintaining (1.0, 0)-DP. In federated learning tasks, GSQ-FL improves accuracy by an average 11.52% over DP-FedPAQ across MNIST and FashionMNIST under non-IID conditions. Additionally, GSQ-FL outperforms DP-FedPAQ by 16.54% on CIFAR-10 and 8.7% on FEMNIST.
{"title":"Randomized Quantization for Privacy in Resource Constrained Machine Learning at-the-Edge and Federated Learning","authors":"Ce Feng;Parv Venkitasubramaniam","doi":"10.1109/TMLCN.2025.3550119","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3550119","url":null,"abstract":"The increasing adoption of machine learning at the edge (ML-at-the-edge) and federated learning (FL) presents a dual challenge: ensuring data privacy as well as addressing resource constraints such as limited computational power, memory, and communication bandwidth. Traditional approaches typically apply differentially private stochastic gradient descent (DP-SGD) to preserve privacy, followed by quantization techniques as a post-processing step to reduce model size and communication overhead. However, this sequential framework introduces inherent drawbacks, as quantization alone lacks privacy guarantees and often introduces errors that degrade model performance. In this work, we propose randomized quantization as an integrated solution to address these dual challenges by embedding randomness directly into the quantization process. This approach enhances privacy while simultaneously reducing communication and computational overhead. To achieve this, we introduce Randomized Quantizer Projection Stochastic Gradient Descent (RQP-SGD), a method designed for ML-at-the-edge that embeds DP-SGD within a randomized quantization-based projection during model training. For federated learning, we develop Gaussian Sampling Quantization (GSQ), which integrates discrete Gaussian sampling into the quantization process to ensure local differential privacy (LDP). Unlike conventional methods that rely on Gaussian noise addition, GSQ achieves privacy through discrete Gaussian sampling while improving communication efficiency and model utility across distributed systems. Through rigorous theoretical analysis and extensive experiments on benchmark datasets, we demonstrate that these methods significantly enhance the utility-privacy trade-off and computational efficiency in both ML-at-the-edge and FL systems. RQP-SGD is evaluated on MNIST and the Breast Cancer Diagnostic dataset, showing an average 10.62% utility improvement over the deterministic quantization-based projected DP-SGD while maintaining (1.0, 0)-DP. In federated learning tasks, GSQ-FL improves accuracy by an average 11.52% over DP-FedPAQ across MNIST and FashionMNIST under non-IID conditions. Additionally, GSQ-FL outperforms DP-FedPAQ by 16.54% on CIFAR-10 and 8.7% on FEMNIST.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"395-419"},"PeriodicalIF":0.0,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10919124","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143645337","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 : 2025-03-06DOI: 10.1109/TMLCN.2025.3567370
Seda Dogan-Tusha;Faissal El Bouanani;Marwa Qaraqe
Federated Learning (FL) has attracted the interest of researchers since it hinders inefficient resource utilization by developing a global learning model based on local model parameters (LMP). This study introduces a novel optimal stopping theory (OST) based online node selection scheme for low complex and multi-parameter FL procedure in IoT networks. Global model accuracy (GMA) in FL depends on the accuracy of the LMP received by the central entity (CE). It is therefore essential to choose trusty nodes to guarantee a certain level of global model accuracy without inducing additional system complexity. For this reason, the proposed technique in this study utilizes the secretary problem (SP) approach as an OST to perform node selection considering both received signal strength (RSS) and local model accuracy (LMA) of available nodes. By leveraging the SP, the proposed technique employs a stopping rule that maximizes the probability of selecting the node with the best quality, and thereby avoids testing all candidate nodes. To this end, this work provides a mathematical framework for maximizing the selection probability of the best node amongst candidate nodes. Specifically, the developed framework has been used to calculate the weighting coefficients of the RSS and LMA to define the node quality. Comprehensive analysis and simulation results illustrate that the OST based proposed technique outperforms state-of-the-art methods including the random node selection and the offline node selection (exhaustive search methods) in terms of GMA and computational complexity, respectively.
{"title":"Optimal Stopping Theory-Based Online Node Selection in IoT Networks for Multi-Parameter Federated Learning","authors":"Seda Dogan-Tusha;Faissal El Bouanani;Marwa Qaraqe","doi":"10.1109/TMLCN.2025.3567370","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3567370","url":null,"abstract":"Federated Learning (FL) has attracted the interest of researchers since it hinders inefficient resource utilization by developing a global learning model based on local model parameters (LMP). This study introduces a novel optimal stopping theory (OST) based online node selection scheme for low complex and multi-parameter FL procedure in IoT networks. Global model accuracy (GMA) in FL depends on the accuracy of the LMP received by the central entity (CE). It is therefore essential to choose trusty nodes to guarantee a certain level of global model accuracy without inducing additional system complexity. For this reason, the proposed technique in this study utilizes the secretary problem (SP) approach as an OST to perform node selection considering both received signal strength (RSS) and local model accuracy (LMA) of available nodes. By leveraging the SP, the proposed technique employs a stopping rule that maximizes the probability of selecting the node with the best quality, and thereby avoids testing all candidate nodes. To this end, this work provides a mathematical framework for maximizing the selection probability of the best node amongst candidate nodes. Specifically, the developed framework has been used to calculate the weighting coefficients of the RSS and LMA to define the node quality. Comprehensive analysis and simulation results illustrate that the OST based proposed technique outperforms state-of-the-art methods including the random node selection and the offline node selection (exhaustive search methods) in terms of GMA and computational complexity, respectively.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"659-676"},"PeriodicalIF":0.0,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10988901","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144100011","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 massive MIMO systems, achieving optimal end-to-end transmission encompasses various aspects such as power control, modulation schemes, path selection, and accurate channel estimation. Nonetheless, optimizing resource allocation remains a significant challenge. In path selection, the direct link is a straightforward link between the transmitter and the receiver. On the other hand, the indirect link involves reflections, diffraction, or scattering, often due to interactions with objects or obstacles. Relying exclusively on one type of link can lead to suboptimal and limited performance. Link management (LM) is emerging as a viable solution, and accurate channel estimation provides essential information to make informed decisions about transmission parameters. In this paper, we study LM and channel estimation that flexibly adjust the transmission ratio of direct and indirect links to improve generalization, using a denoising variational autoencoder with attention modules (DVAE-ATT) to enhance sum rate. Our experiments show significant improvements in IRS-assisted millimeter-wave MIMO systems. Incorporating LM increased the sum rate and reduced MSE by approximately 9%. Variational autoencoders (VAE) outperformed traditional autoencoders in the spatial domain, as confirmed by heatmap analysis. Additionally, our investigation of DVAE-ATT reveals notable differences in the temporal domain with and without attention mechanisms. Finally, we analyze performance across varying numbers of users and ranges. Across various distances—5m, 15m, 25m, and 35m—performance improvements averaged 6%, 11%, 16%, and 22%, respectively.
{"title":"Paths Optimization by Jointing Link Management and Channel Estimation Using Variational Autoencoder With Attention for IRS-MIMO Systems","authors":"Meng-Hsun Wu;Hong-Yunn Chen;Ta-Wei Yang;Chih-Chuan Hsu;Chih-Wei Huang;Cheng-Fu Chou","doi":"10.1109/TMLCN.2025.3547689","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3547689","url":null,"abstract":"In massive MIMO systems, achieving optimal end-to-end transmission encompasses various aspects such as power control, modulation schemes, path selection, and accurate channel estimation. Nonetheless, optimizing resource allocation remains a significant challenge. In path selection, the direct link is a straightforward link between the transmitter and the receiver. On the other hand, the indirect link involves reflections, diffraction, or scattering, often due to interactions with objects or obstacles. Relying exclusively on one type of link can lead to suboptimal and limited performance. Link management (LM) is emerging as a viable solution, and accurate channel estimation provides essential information to make informed decisions about transmission parameters. In this paper, we study LM and channel estimation that flexibly adjust the transmission ratio of direct and indirect links to improve generalization, using a denoising variational autoencoder with attention modules (DVAE-ATT) to enhance sum rate. Our experiments show significant improvements in IRS-assisted millimeter-wave MIMO systems. Incorporating LM increased the sum rate and reduced MSE by approximately 9%. Variational autoencoders (VAE) outperformed traditional autoencoders in the spatial domain, as confirmed by heatmap analysis. Additionally, our investigation of DVAE-ATT reveals notable differences in the temporal domain with and without attention mechanisms. Finally, we analyze performance across varying numbers of users and ranges. Across various distances—5m, 15m, 25m, and 35m—performance improvements averaged 6%, 11%, 16%, and 22%, respectively.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"381-394"},"PeriodicalIF":0.0,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10909334","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143583165","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：