IEEE Transactions on Machine Learning in Communications and Networking最新文献

{"title":"Learning to Optimize Voltage Level for Energy-Efficient Radios","authors":"Cecilia Nyberg;Elin Stiebe;Thomas Lejon;Linus Aronsson;Morteza Haghir Chehreghani","doi":"10.1109/TMLCN.2026.3711568","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3711568","url":null,"abstract":"Reducing energy waste in cellular radio systems is critical for lowering the carbon footprint of wireless communication networks. In current deployments, radio power amplifiers (PAs) typically operate at a fixed high supply voltage regardless of physical resource block utilization (PRB-U), leading to unnecessary energy consumption. This work investigates a machine learning-based approach on real field data for dynamically switching the PA supply voltage between two hardware-supported levels, thereby adapting the available radio-frequency (RF) output power capacity to predicted traffic demand. While many prior approaches focus on network-level control, the proposed method performs voltage optimization at the level of individual radio units. Eight models are evaluated under two task formulations: 1) classification of voltage levels and 2) regression of PRB-U followed by mapping to voltage states. The results show that machine learning can achieve substantial energy savings, although effectiveness varies across radios. Underestimations, defined as instances where the predicted voltage is lower than required, are minimized more effectively by classification models than by regression approaches, resulting in a more reliable trade-off between energy efficiency and connectivity. A customized Feedforward Neural Network classifier and a Random Forest model perform best, outperforming both statistical baselines and more complex temporal architectures, achieving F1 scores of 0.35 and 0.29 respectively while reducing energy consumption by over 12%. In contrast, a baseline ARIMA model shows poor predictive performance with an F1 score of 0.09 despite similar energy savings. Explainability analysis using SHAP indicates that current PRB-U is the most influential feature, while the superior performance of nonlinear models suggests that linear utilization-based approaches are insufficient for reliable switching decisions. These results demonstrate the potential of learning-based PA voltage control as a new mechanism for improving the energy efficiency of cellular radio systems while maintaining service reliability.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"1199-1226"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11602085","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148494637","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}

{"title":"Privacy-Preserving Distributed Online Learning With Zeroth-Order Feedback Over Directed Networks","authors":"Qingguo Lü;Chenglong He;Jingxin Liu;Harish Garg;Yingxu Wang;Jin-Song Dong","doi":"10.1109/TMLCN.2026.3737073","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3737073","url":null,"abstract":"This paper investigates a zeroth-order online optimization problem over a directed network, where each node can only access the function value of its local loss after receiving the decision variable, and information exchange among nodes may incur potential privacy leakage risks. To address these challenges, we propose an efficient chaos-encrypted distributed dual averaging algorithm with state decomposition, termed CEDDA-SD. The CEDDA-SD algorithm employs a zeroth-order gradient oracle to effectively overcome the absence of explicit gradient information. Meanwhile, a multi-step accelerated method is incorporated to enhance convergence performance. Furthermore, a combination of state decomposition and chaos encryption strategies is introduced to defend against prevalent privacy inference attacks. Rigorous theoretical analysis establishes that the CEDDA-SD algorithm achieves sublinear expected regret while guaranteeing privacy preservation. Simulations and field-programmable gate array (FPGA) experiments validate the effectiveness of the proposed algorithm and demonstrate its practical advantages in terms of privacy preservation.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"1513-1526"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11704845","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149131436","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}

{"title":"Robust Unsupervised Learning for Wireless Resource Allocation Under Channel Distribution Shift","authors":"Shengjie Liu;Chenyang Yang","doi":"10.1109/TMLCN.2026.3736371","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3736371","url":null,"abstract":"Unsupervised learning is increasingly used to train deep neural networks (DNNs) for optimizing wireless resource allocation, avoiding expensive label generation. In practical deployments, however, the test distribution of wireless channels often deviates from training owing to the diverse scenarios and configurations of wireless systems, leading to degradation in out-of-distribution (OOD) generalization performance. In this paper, we develop offline, model-agnostic training methods that improve the robustness of DNNs using only single-source unlabeled data. The idea is to refine a trained DNN using progressively identified worst-case samples out of the training distribution, inspired by the principle of distributionally robust optimization (DRO). To mitigate the OOD deterioration of a DNN, we formulate DRO for optimizing on the “true” worst-case distribution in the context of unsupervised learning, which maximizes the optimality gap (relative to the unknown optimum) rather than the loss value itself. We then provide its surrogate adversarial training problem and propose two training methods: 1) Unsupervised Adversarial Training (UAT) for the resource allocation problems whose constraints can be enforced by output projections and 2) Extended UAT (EUAT) for problems with input-dependent constraints (e.g., quality-of-service requirements), where the robustness must capture both objective degradation and constraint violations. In both methods, the gradients required to construct adversarial examples are obtained using the envelope theorem, avoiding access to the optimal policy. Extensive simulations on hybrid precoding and cell-free power allocation — across 3GPP, DeepMIMO, and real-world channels in different scenarios — demonstrate consistent OOD performance gains and improved constraint satisfaction for multiple kinds of DNNs with no added inference cost.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"1569-1587"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11703863","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149188819","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}

{"title":"Generalization in Reinforcement Learning for Radio Access Networks","authors":"Burak Demirel;Yu Wang;Cristian Tatino;Pablo Soldati","doi":"10.1109/TMLCN.2026.3655335","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3655335","url":null,"abstract":"Modern radio access networks (RANs) operate in highly dynamic and heterogeneous environments, where hand-tuned, rule-based radio resource management (RRM) algorithms frequently underperform. While reinforcement learning (RL) can surpass these heuristics in constrained scenarios, the unpredictable nature of radio channels and the diversity of cell deployments introduce substantial generalization challenges. Data-driven policies often overfit to their training distributions, leading to degraded performance when applied to unseen conditions. To address this, we propose a generalization-focused RL framework for RAN control that: (i) robustly reconstructs dynamically varying states from partial and noisy observations, while encoding static and semi-static information—such as radio nodes, cell attributes, and their topology—through graph representations; (ii) applies extensive domain randomization to broaden the training distribution; and (iii) distributes data generation across multiple actors while centralizing training in a cloud-compatible architecture aligned with O-RAN principles. Although training generalizable policies increases computational and data-management complexity, our distributed design mitigates these costs by scaling data collection and training across heterogeneous network conditions. Applied to downlink link adaptation (LA) across five <inline-formula> <tex-math>$5^{text {th}}$ </tex-math></inline-formula> Generation (5G) benchmarks, the resulting generalized RL policy achieves approximately 10% higher average cell throughput and spectral efficiency than the state-of-the-art LA baseline in full-buffer multiple input multiple output (MIMO) and massive MIMO (mMIMO) scenarios, and approximately 20% under high-mobility conditions. Furthermore, it matches the performance of specialized RL policies in full-buffer traffic while delivering up to <inline-formula> <tex-math>$4times $ </tex-math></inline-formula> and <inline-formula> <tex-math>$2times $ </tex-math></inline-formula> throughput gains in enhanced mobile broadband (eMBB) and mixed-traffic benchmarks, respectively. In larger deployments with nine cells, graph attention network (GAT) models achieve an additional 30% throughput gain over multi-layer perceptron (MLP) baselines, underscoring the effectiveness of attention-based architectures in larger network settings. These performance gains, combined with the scalable and distributed learning architecture, provide a practical foundation for AI-native <inline-formula> <tex-math>$6^{text {th}}$ </tex-math></inline-formula> Generation (6G) RANs, enabling the deployment of a single, generalizable RL agent that operates network-wide and consistently outperforms traditional heuristics.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"351-372"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11358408","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148494642","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}

{"title":"Guest Editorial Parting Words From Inaugural Editor-in-Chief","authors":"Walid Saad","doi":"10.1109/TMLCN.2026.3713004","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3713004","url":null,"abstract":"","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"1279-1280"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11643500","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148749140","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}

{"title":"Convolutional Neural Network-Based Data Distribution Routing Framework for Service Function Chaining for ROS-Based Industrial Automation","authors":"Amir Hoseein Ghorab;Mohammed Abuibaid;Aysun Aslan Saruhan;Marc St-Hilaire","doi":"10.1109/TMLCN.2026.3722593","DOIUrl":"https://doi.org/10.1109/TMLCN.2026.3722593","url":null,"abstract":"With continuous advancements in robotics and networking, the adoption of smart industrial machines for automation has grown significantly. Leveraging the Robot Operating System version 2 (ROS-2) and its topic-based data distribution mechanism, these machines are typically controlled by software deployed in the cloud. This control software is structured as a chain of ROS-2 nodes, with each node responsible for managing a specific function or component of the industrial machine. By applying the concept of Service Function Chaining (SFC), these ROS-2 node chains can be deployed in a distributed manner across cloud infrastructure. An SDN controller is employed to steer ROS-2 traffic efficiently along the sequence of nodes. As the number of industrial machines increases in large-scale factories, the number of corresponding ROS-2 node chains also grows. This trend underscores the need to reuse common ROS-2 nodes across multiple chains to optimize the utilization of cloud resources. However, the default topic-based communication model of ROS-2 is not inherently compatible with such reuse. Sharing common ROS-2 nodes across different chains becomes challenging, thereby hindering efficient resource management in cloud environments. To address this limitation, this paper proposes a Data Distribution Routing (DDR) framework powered by a Spatial Attention Convolutional Neural Network (SA-CNN). The DDR is capable of classifying and steering ROS-2 traffic originating from shared nodes to their appropriate service chains. Furthermore, it supports the NetConf protocol to communicate with the Software-Defined Networking (SDN) controller. Our evaluation includes both neural network performance and system-level metrics. The SA-CNN outperforms Vanilla CNN and CNN+LSTM in classification accuracy and efficiency. Additionally, the DDR shows favorable resource usage, latency, and throughput under real-world workloads.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"1371-1388"},"PeriodicalIF":4.9,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11649542","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148781264","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}

{"title":"Clustering-Assisted Deep Reinforcement Learning for Joint Trajectory Design and Resource Allocation in Two-Tier-Cooperated UAVs Communications","authors":"Shujun Zhao;Simeng Feng;Chao Dong;Xiaojun Zhu;Qihui Wu","doi":"10.1109/TMLCN.2025.3647806","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3647806","url":null,"abstract":"Considering their high mobility and relatively low cost, uncrewed aerial vehicles (UAVs) equipped with mobile base stations are regarded as a potential technological approach. However, the dual pressures of limited onboard resources of UAVs and the demand for high-quality services in dynamic low-altitude applications jointly form a bottleneck for system performance. Although multi-UAVs communication networks can provide higher system performance through coordinated deployment, the challenges of cooperation and competition among UAVs, as well as more complex optimization problems, significantly increase costs and pose formidable challenges. To overcome the challenges of low coordination efficiency and intense resource competition among multiple UAVs, and to ensure the timely and efficient satisfaction of ground users (GUs) communication service demands, this paper conceives a centralized-controlled two-tier-cooperated UAVs communication network. The network comprises a central UAV (C-UAV) tier as control center and a marginal UAV (M-UAV) tier to serve GUs. In response to the increasingly dynamic and complex scenarios, along with the challenge of insufficient generalization ability in Deep Reinforcement Learning (DRL) algorithms, we propose a clustering-assisted dual-agent soft actor critic (CDA-SAC) algorithm for trajectory design and resource allocation, aiming to maximize the fair energy efficiency of the system. Specifically, by integrating a clustering-matching method with a dual-agent strategy, the proposed CDA-SAC algorithm achieves significant improvements in generalization ability and exploration capability. Simulation results demonstrate that the proposed CDA-SAC algorithm can be deployed without retraining in scenarios with different numbers of GUs. Furthermore, the CDA-SAC algorithm outperforms both the multi-UAV scenarios based on the MADDPG algorithm and the FDMA scheme in terms of fairness and total energy efficiency.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"178-197"},"PeriodicalIF":0.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11313631","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886662","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}

{"title":"Data-Driven Cellular Mobility Management Via Bayesian Optimization and Reinforcement Learning","authors":"Mohamed Benzaghta;Sahar Ammar;David López-Pére;Basem Shihada;Giovanni Geraci","doi":"10.1109/TMLCN.2025.3647807","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3647807","url":null,"abstract":"Mobility management in cellular networks faces increasing complexity due to network densification and heterogeneous user mobility characteristics. Traditional handover (HO) mechanisms, which rely on predefined parameters such as A3-offset and time-to-trigger (TTT), often fail to optimize mobility performance across varying speeds and deployment conditions. Fixed A3-offset and TTT configurations either delay HOs, increasing radio link failures (RLFs), or accelerate them, leading to excessive ping-pong effects. To address these challenges, we propose two distinct data-driven mobility management approaches leveraging high-dimensional Bayesian optimization (HD-BO) and deep reinforcement learning (DRL). While HD-BO optimizes predefined HO parameters such as A3-offset and TTT, DRL provides a parameter-free alternative by allowing an agent to select serving cells based on real-time network conditions. We systematically compare these two approaches in real-world site-specific deployment scenarios (employing Sionna ray tracing for site-specific channel propagation modeling), highlighting their complementary strengths. Results show that both HD-BO and DRL outperform 3GPP set-1 (TTT of 480 ms and A3-offset of 3 dB) and set-5 (TTT of 40 ms and A3-offset of −1 dB) benchmarks. We augment HD-BO with transfer learning so it can generalize across a range of user speeds. Applying the same transfer-learning strategy to the DRL method reduces its training time by a factor of 2.5 while preserving optimal HO performance, showing that it adapts efficiently to the mobility of aerial users such as UAVs. Simulations further reveal that HD-BO remains more sample-efficient than DRL, making it more suitable for scenarios with limited training data.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"228-244"},"PeriodicalIF":0.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11313634","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929689","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}

{"title":"Transforming Indoor Localization: Advanced Transformer Architecture for NLOS Dominated Wireless Environments With Distributed Sensors","authors":"Saad Masrur;Jung-Fu Cheng;Atieh R. Khamesi;İsmail Güvenç","doi":"10.1109/TMLCN.2025.3647376","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3647376","url":null,"abstract":"Indoor localization in challenging non-line-of-sight (NLOS) environments often leads to poor accuracy with traditional approaches. Deep learning (DL) has been applied to tackle these challenges; however, many DL approaches overlook computational complexity, especially for floating-point operations (FLOPs), making them unsuitable for resource-limited devices. Transformer-based models have achieved remarkable success in natural language processing (NLP) and computer vision (CV) tasks, motivating their use in wireless applications. However, their use in indoor localization remains nascent, and directly applying Transformers for indoor localization can be both computationally intensive and exhibit limitations in accuracy. To address these challenges, in this work, we introduce a novel tokenization approach, referred to as Sensor Snapshot Tokenization (SST), which preserves variable-specific representations of power delay profile (PDP) and enhances attention mechanisms by effectively capturing multi-variate correlation. Complementing this, we propose a lightweight Swish-Gated Linear Unit-based Transformer (L-SwiGLU-T) model, designed to reduce computational complexity without compromising localization accuracy. Together, these contributions mitigate the computational burden and dependency on large datasets, making Transformer models more efficient and suitable for resource-constrained scenarios. Experimental results on simulated and real-world datasets demonstrate that SST and L-SwiGLU-T achieve substantial accuracy and efficiency gains, outperforming larger Transformer and CNN baselines by over 40% while using significantly fewer FLOPs and training samples.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"161-177"},"PeriodicalIF":0.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11313538","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145886663","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}

{"title":"Explainable Multi-Agent Reinforcement Learning for Extended Reality Codec Adaptation","authors":"Pedro Enrique Iturria-Rivera;Raimundas Gaigalas;Medhat Elsayed;Majid Bavand;Yigit Ozcan;Melike Erol-Kantarci","doi":"10.1109/TMLCN.2025.3646125","DOIUrl":"https://doi.org/10.1109/TMLCN.2025.3646125","url":null,"abstract":"Extended Reality (XR) services are set to transform applications over <inline-formula> <tex-math>${mathbf {5}}^{th}$ </tex-math></inline-formula> and <inline-formula> <tex-math>${mathbf {6}}^{th}$ </tex-math></inline-formula> generation wireless networks, delivering immersive experiences. Concurrently, Artificial Intelligence (AI) advancements have expanded their role in wireless networks, however, trust and transparency in AI remain to be strengthened. Thus, providing explanations for AI-enabled systems can enhance trust. We introduce Value Function Factorization (VFF)-based Explainable (X) Multi-Agent Reinforcement Learning (MARL) algorithms, explaining reward design in XR codec adaptation through reward decomposition. We contribute four enhancements to XMARL algorithms. Firstly, we detail architectural modifications to enable reward decomposition in VFF-based MARL algorithms: Value Decomposition Networks (VDN), Mixture of Q-Values (QMIX), and Q-Transformation (Q-TRAN). Secondly, inspired by multi-task learning, we reduce the overhead of vanilla XMARL algorithms. Thirdly, we propose a new explainability metric, Reward Difference Fluctuation Explanation (RDFX), suitable for problems with adjustable parameters. Lastly, we propose adaptive XMARL, leveraging network gradients and reward decomposition for improved action selection. Simulation results indicate that, in XR codec adaptation, the Packet Delivery Ratio reward is the primary contributor to optimal performance compared to the initial composite reward, which included delay and Data Rate Ratio components. Modifications to VFF-based XMARL algorithms, incorporating multi-headed structures and adaptive loss functions, enable the best-performing algorithm, Multi-Headed Adaptive (MHA)-QMIX, to achieve significant average gains over the Adjust Packet Size baseline up to 10.7%, 41.4%, 33.3%, and 67.9% in XR index, jitter, delay, and Packet Loss Ratio (PLR), respectively.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"4 ","pages":"245-264"},"PeriodicalIF":0.0,"publicationDate":"2025-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11303975","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145929688","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}