最新刊期
卷 8 ， 期 4 ， 2024
- 摘要：In the context of the rapid development of Internet of things (IoT) technology, IoT devices faced challenges in running complex artificial intelligence (AI) algorithms, especially deep learning models, due to the limitations of computing power, storage space, communication bandwidth, and battery life. Model pruning technology could effectively reduce computation and storage requirements by reducing redundant parameters in neural networks without impairing the performance of AI models. This technique was extremely suitable for optimising AI models deployed on IoT devices. Firstly, two typical model pruning techniques-structured pruning and unstructured pruning, which were currently popular and suitable for different application scenarios, were reviewed. Secondly, the diverse applications of these methods in IoT environments were analysed in detail. Finally, the limitations of the current model pruning were discussed in detail in the light of the latest research results, and the future development direction of model pruning methods in IoT was outlooked.关键词：IoT;resource constraints;model pruning;AI;deep learning310|2423|0更新时间：2025-01-20
- 摘要：The synchronous training mechanism of traditional federated learning was not suitable for dynamic vehicle computing power network scenarios, and lacked effective detection mechanisms under the threat of malicious vehicle attacks. To address the above issues, an asynchronous robust federated learning method was proposed, which achieves vehicle data privacy protection while improving the efficiency of model collaborative training through asynchronous execution of federated learning processes between vehicles. Secondly, a model selection method was designed, and potential malicious model detection and vehicle reputation evaluation methods are proposed to further enhance the robustness of the system. Then, the safety of the proposed method was analyzed in detail from a probabilistic perspective, providing a theoretical basis for optimizing various parameters. Finally, the simulation results show that this method can achieve efficient asynchronous federated learning while having good robustness.关键词：vehicle computing power network;federated learning;robustness;asynchronous learning277|521|0更新时间：2025-01-20
- 摘要：Multi-sensor systems integrate diverse sensor data to achieve comprehensive and accurate environmental perception. However, how to effectively fuse heterogeneous data and realize the efficiency of real-time processing is still a hot and difficult issue in current research. Therefore, focusing on data fusion and arithmetic optimization of multi-source heterogeneous sensors, an innovative solution was proposed. Firstly, a data fusion system based on master-slave architecture was designed to solve the problem of multi-source heterogeneous data processing. Secondly, a three-layer "cloud-edge-end" architecture was implemented, leveraging edge servers to offload computational pressure from cloud servers, optimizing task scheduling strategies, and enabling coordinated management of network and computing resources. Finally, the delay and energy consumption requirements of tasks were modeled, and the optimization problem of minimizing system cost was constructed under resource constraints, which was transformed into Markov decision process (MDP) and solved with deep deterministic policy gradient (DDPG) algorithm. Simulation experiments show that the proposed architecture and scheduling algorithm exhibit excellent performance in reducing both latency and energy consumption, and provide a new idea for efficient data fusion and arithmetic optimization in multi-sensor systems.关键词：multi source heterogeneous data;data fusion;sensor;arithmetic optimization792|689|0更新时间：2025-01-20
- 摘要：With the rapid advancement of intelligent businesses, the pre-existing relationship between traditional network architectures and computing capabilities has made it difficult to meet the current demands, making the implementation of computing-network convergence inevitable. Under the new computing power network framework brought about by the convergence of computing networks, efficient and intelligent resource scheduling strategy has become a key link to improve user experience. However, the existing resource scheduling algorithms have a single optimization objective and cannot meet the differentiated business needs of multi-tenants. To this end, a Multi objective deep reinforcement learning resource scheduling (MODRLRS) was proposed to call the computing resources and network resources in the computing power network. The algorithm performs multi-objective scheduling optimization of computing network resources by constructing a Pareto optimal solution set to meet the personalized business needs of different tenants. Simulation experimental results show that compared with other multi-objective resource scheduling algorithms, the proposed algorithm improves the request acceptance rate by 4.9% and the compliant delay request rate by 4.78%, which can flexibly adapt to the unique requirements of various computing services.关键词：integration of computing and networking;computing power network;resource scheduling;multi objective optimization;deep reinforcement learning551|326|0更新时间：2025-01-20
- 摘要：Body area network (BAN) is a key technology of the medical Internet of things for personal health monitoring. Integrated with edge computing, it realizes real-time monitoring of physiological data, emergency warning, and intelligent treatment and diagnosis. However, the quality of service (QoS) requirements of the computing tasks in BAN varie with the urgency of the sensing data. The existing resource allocation methods in edge computing network are difficult to efficiently and flexibly support dynamic QoS of multi-source heterogeneous tasks in BAN. A dynamic QoS-aware stochastic optimization problem on computation offloading decisions and edge computing resource allocation was studied. Firstly, considering the Markov nature of multi-source task priorities and channel state changes in BAN, the original stochastic optimization problem was transformed into an infinite horizon Markov decision process problem. Then, a multi-source task priority sequence for each BAN was constructed and an online decision-making method that integrated proximal policy optimization (PPO) was proposed for task offloading and computing resource allocation. The simulation results show that the proposed optimization scheme outperforms existing baseline methods, effectively meeting the dynamic priority requirements of tasks in BAN and reducing the energy consumption as well as the average delay required for task completion.关键词：medical Internet of things;edge computing;resource management;QoS236|823|0更新时间：2025-01-20
- 摘要：Federated computing power Internet of things (IoT) is designed to deeply integrate computing power with IoT resources, facilitating the efficient utilization of vast and ubiquitously dispersed IoT data and heterogeneous resources through federated learning. Faced with the threats of emerging privacy attacks, e.g., model inversion attacks and gradient leakage attacks, the academic and industrial communities have widely investigated and applied differential privacy (DP) as an effective privacy protection technique. However, two severe challenges have not been taken into account in the existing DP budget settings, i.e., data heterogeneity issue of local computing power nodes and the fairness of privacy budget allocation, which lead to a significant loss in model accuracy. Therefore, an adaptive optimization scheme for privacy budget was proposed in federated computing power IoT, which was called federated learning based on Cramér-Rao lower bound differential privacy (FedCDP). In specific, to adaptively adjust privacy budgets, the privacy budget estimates for edge computing power nodes based on the Cramér-Rao lower bound theory were analyzed. Furthermore, by assessing the similarity between the local model and the aggregated model, as well as their respective privacy budget proportions, the global contribution of each node was determined, which was used to fairly, also in real time, optimize and adjust the privacy budget settings in conjunction with the estimated privacy budget. Through rigorous theoretical analysis, FedCDP achieves ε-DP for local models, and ensures the convergence of the global model. Experimental results on multiple public datasets show that the proposed scheme improves the accuracy of the global model by up to 10.19% under the premise of satisfying the same privacy protection requirements.关键词：federated computing power IoT;differential privacy;privacy budget;adaptive optimization525|365|0更新时间：2025-01-20
- 摘要：Computing power Internet of things (CPIoT) integrates Internet of things (IoT) devices with substantial computational resources to support data-intensive tasks, facilitating intelligent decision-making. Within the context of privacy protection requirements for CPIoT, federated learning (FL) that is a distributed learning technique upholds data privacy, and offers a novel approach to addressing data silos for executing complex training tasks, and training large models. Although researchers have been committed to develop more mature federated learning systems to adapt to the CPIoT environment, current research lacks in-depth exploration of the strengths and limitations, technical features and differences, and support and applicability of federated learning system design techniques. Firstly, the most influential federated learning systems in the industry were studied, including open-source frameworks and benchmarking platforms. The system design differences in various technical dimensions of CPIoT in an in-depth comparison were analyzed. Detailed criteria and recommendations for selecting open-source frameworks and benchmarking platforms in the CPIoT environment were established, so that developers could efficiently choose the most suitable frameworks and platforms. Seeondly, various experiments for selecting federated learning systems and building complete systems were presented in multiple CPIoT scenarios, to assist developers in better realizing federated learning applications by utilizing the aforementioned technologies. Finally, the current state of standardization and development challenges in the field of federated learning system design were summarized, and future development prospects were discussed. The purpose is to provide a comprehensive overview of FL systems and the design research progress, serving as a reference for the deep integration of CPIoT and FL networks and offering insights for future research.关键词：CPIoT;FL;open-source framework;benchmarking platform;computing paradigm330|339|0更新时间：2025-01-20
- 摘要：In vehicle computing, the intelligent vehicle which has strong computing capability and abundant sensing devices provides services for users. Many sensing devices can provide services for users without the limits of time and place. Intelligent vehicles have large amount computing and sensing resources, where computing resources are used individually by users while sensing resources can be shared by multiple users. According to the characteristics of intelligent vehicles, a new resource allocation model based on resource sharing was proposed. A resource sharing allocation method based on dwarf mongoose optimization was proposed. A repairing algorithm was proposed to transform infeasible solutions into feasible solutions. A new solution generation algorithm based on the random and greedy strategy was proposed to address the problem of the dwarf mongoose optimization algorithm being prone to getting stuck in local optima, to improve the convergence speed and obtain the optimal solution. The experimental results show that the proposed strategy performs well in different allocation environments and is adaptable.关键词：vehicle computing;dwarf mongoose optimization algorithm;resource sharing;resource allocation125|380|0更新时间：2025-01-20
-
A fair multi-resource allocation mechanism for time-varying discrete jobs with placement constraints
摘要：A key issue in resource sharing in cloud computing is how to fairly and efficiently allocate the multi-resources to users with dynamic demand. Multi-resource fair allocation in a cloud computing system usually faces problems, such as subdividing the minimum granularity of users' resource requirements, and the mismatch between task requirements and server configurations. Most of the existing mechanisms for multi-resource fair allocation are based on the ideal assumption that the task demands of user are infinitely divisible or that the task execution and server configuration are matched, which makes it difficult to guarantee that the allocation is feasible. By analyzing the characteristics of time-varying indivisible task demands and task placement constraints, a time-varying task share fairness allocation mechanism based on cumulative task share fairness was designed to ensure the fairness and efficiency of resource allocation. Theoretical analysis shows that the TV-TSF mechanism satisfies the sharing incentive, envy-freeness up to one item, and Pareto optimal properties. Simulation results based on the Alibaba cluster dataset show that, compared with the existing fair allocation mechanisms, the TV-TSF mechanism proposed can effectively reduce the waiting time, job queuing time, and job completion time of users.关键词：dynamic multi-resource allocation;indivisible task demand;task placement constraint;cumulative task share fairness291|220|0更新时间：2025-01-20
Topic: Computing Power Internet of Things
- 摘要：Passive backscatter based ambient Internet of things (AIoT) was an important development direction for the future of IoT and currently attracted extensive attentions. In practical applications of AIoT, there existed strong self-interference caused by phase noise and spurs, which brought about new challenges for channel estimation. Therefore, an iterative channel estimator considering phase noise and spurs were designed for the AIoT system with two nodes. Specifically, the estimator was based on the least squares method and complex exponential basis expansion model (CE-BEM), and used iteration to improve estimation accuracy. The Cramér-Rao lower bound (CRLB) of channel estimation parameters was also derived to evaluate the theoretical limit of the estimation accuracy. Finally, simulation results were provided to corroborate the proposed studies.关键词：backscatter communication;channel estimation;complex exponential basis expansion model;Cramér-Rao lower bound;internet of things;phase noise;spurs164|200|0更新时间：2025-01-20
- 摘要：The issue of channel estimation for a double intelligent reflecting surface (IRS) assisted millimeter wave multiple-input multiple-output (MIMO) system was addressed and a channel estimation scheme based on tensor decomposition and manifold optimization was proposed. Specifically, a tensor model was constructed based on the high-dimensional features of received signals, and the objective function of the channel estimation problem was formulated based on the Tucker2 decomposition of the tensor. Then, the channel estimation problem was decomposed into multiple sub-problems using alternating optimization theory, providing feasible solutions for estimating the channel of each hop in the double IRS scenario. Finally, considering the low-rank characteristics of the millimeter wave channel itself, each channel estimation sub-problem was transformed into an optimization problem on the complex fixed-rank matrix manifold, and a manifold optimization-based alternating channel estimation scheme was proposed by leveraging the advantages of fixed-rank manifold optimization in solving rank-constrained optimization problems. Unlike traditional schemes, the proposed scheme takes into account the low-rank characteristics of millimeter wave channels, accurately describes the channels, and effectively handles fixed-rank constraints using manifold optimization theory, thus improving the accuracy of channel estimation. Simulation results show that the proposed channel estimation scheme outperforms existing reference schemes in terms of estimation performance in different scenarios.关键词：millimeter wave MIMO system;double IRS;channel estimation;manifold optimization;tensor decomposition148|327|0更新时间：2025-01-20
- 摘要：The problem of region of interest (RoI) extraction and transmission of video frames captured in edge-assisted unmanned aerial vehicle (UAV) systems was investigated to improve the inference performance of patrolling tasks. Due to the limited UAV onboard computational resources, a lightweight RoI extraction method based on class activation mapping (CAM) was proposed, which was able to rapidly locate areas containing patrolling targets. Those RoIs were then transmitted to edge servers for further processing. To address the challenges from dynamic UAV trajectories and fluctuating network conditions, the RoIs collected by UAVs were properly choosen through an adaptive RoI box selection algorithm, followed by adaptive configuration of quantization parameters (QP) of video codec, in order to further compress the transmitted data volume. A joint optimization problem was thus formulated for RoI box selection and adaptive coding configuration, which was solved via a heuristic algorithm. Experimental results demonstrate that, the proposed approach can effectively improve the detection accuracy of patrolling tasks, reduce data transmission volume, and significantly lower system latency, indicating great potential in UAV-based patrolling applications.关键词：UAV communication;mobile edge computing;RoI extraction;video coding907|1478|0更新时间：2025-01-20
- 摘要：With the rapid proliferation of Internet of things (IoT) devices, the frequency and intensity of attacks targeting these devices are constantly increasing. Therefore, it's quite important that security mechanisms are continuously updated to ensure the safety of IoT devices. However, as public awareness of privacy grows, many datasets are no longer shared, leading to the emergence of data silos, which hinders the improvement of IoT security. To address this issue, a federated reinforcement learning-based intrusion detection method was proposed, and experiments were conducted using two datasets from the Internet of medical things (IoMT) and Internet of vehicles (IoV) scenarios. Imbalanced traffic sample distributions were designed for each edge agent to simulate a real-world environment, allowing for the evaluation of the detection accuracy and robustness of the global model. Double deep Q-network (DDQN) was employed as the reinforcement learning framework for the edge agents, and the experimental results were evaluated using accuracy, precision, recall, and F1-score. The results demonstrate that the proposed method exhibits strong robustness and detection accuracy.关键词：federated reinforcement learning;intrusion detection;IoT security;IoT333|502|0更新时间：2025-01-20
- 摘要：Synthetic aperture radar (SAR) remote sensing images have been widely applied in military reconnaissance and traffic supervision, owing to their all-weather and all-day abilities. With excellent learning performance, convolutional neural networks are employed in the SAR ship detection algorithms. However, it is difficult to extract features. In practical applications, computing resources and memory space are limited, and high inference speed is required. Therefore, a lightweight attention-based ship detector (LASD) was proposed. A novel linear hybrid attention module was designed which extracted potential ship features from deep-level space by using global channel attention and local spatial attention. A spatial pyramid pooling module based on cross-stage partial connections optimized the quality of multi-scale feature fusion, which replaced the parallel max-pooling group with large kernels with the serial max-poolings with small kernels to improve the inference speed. A novel feature fusion scheme via the local channel attention was suggested which widened the gap between the objects and background noise using local attention during the feature fusion. The results on the public datasets SSDD and LS-SSDD-v1.0 show that LASD achieves the balance of detection precision and inference speed, and is more competitive than the other advanced algorithms.关键词：SAR;ship detection;convolutional neural network;attention mechanism;multi-scale feature fusion294|1164|0更新时间：2025-01-20
- 摘要：To solve the problem of high feedback overhead in a multi-user massive multiple-input multiple-output (MIMO) system assisted by a reconfigurable intelligent surface (RIS) in frequency-division duplexing (FDD) mode, a channel state information (CSI) feedback framework based on manifold learning was proposed. Firstly, the framework achieved initial feedback overhead reduction by simplifying the CSI feedback process. Then, the framework combined the manifold learning to train two set of dictionaries to achieve dimension reduction and reconstruction of incremental CSI. Finally, the original channel was restored at the base station. The simulation results show that the CSI feedback scheme proposed in this paper has lower overhead and complexity than the existing methods in the multi-user and limited scattering environment, and the reconstruction quality is significantly improved.关键词：massive MIMO;FDD;RIS;channel feedback;manifold learning631|324|0更新时间：2025-01-20
Theory and Technology
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：212817 Visits Today：640
0