检索结果-内蒙古大学图书馆

TinyFEL: Communication, computation, and Memory Efficient Tiny Federated Edge Learning via Model Sparse Update

IEEE INTERNET OF THINGS JOURNAL 2025年第7期12卷 8247-8260页

作者： Chen, Qimei Cheng, Han Liang, Yipeng Zhu, Guangxu Li, Ming Jiang, Hao Wuhan Univ Sch Elect Informat Wuhan 430072 Peoples R China Shenzhen Res Inst Big Data Shenzhen 518172 Peoples R China Fiberhome Commun Technol Co Ltd Wuhan 430074 Hubei Peoples R China

Federated edge learning (FEL) is regarded as a promising distributed machine learning paradigm to reduce transmission latency and resources as well as preserve raw data privacy by collaboratively training local deep learning models across multiple edge devices. However, with the development of artificial intelligence (AI) technologies, the size of neural network models grows exponentially with their parameters to meet variable application requirements, which poses significant challenges to the computation, communication, and memory abilities of edge devices. Existing designs typically focus on either communication or computation efficiency without caring each device's memory ability. To deal with the above issues, we first introduce a novel model sparse update enabled tiny FEL (TinyFEL) architecture, which terminates the backpropagation early in local model training processes. Therefore, the proposed TinyFEL can reduce local memory occupation and lessen the communication-and-computation burden. Furthermore, we propose a parameter splitting mechanism instead of transmitting the full model, only a part of updated layers of parameters is transmitted for aggregation, which significantly reduced the communication overheads. Thereafter, we develop a communication and computation latency minimization problem to accelerate the training of TinyFEL. To this end, we theoretically analyze the convergence performance of TinyFEL, which unveils the mathematical relationship among sparse update ratio assignment, device selection, and learning performance. Then, a joint sparse update ratio assignment, device selection, and resource allocation strategy is introduced based on the alternating direction method of multipliers (ADMMs) and block coordinate descent (BCD) algorithms. Numerical results indicate that our proposed TinyFEL can reduce training memory occupation by over 40% than the traditional FEL at the cost of negligible accuracy loss.

关键词： computational modeling Training Performance evaluation Memory management Servers Data models Optimization Numerical models Resource management Mathematical models Communication resources computation resources federated edge learning (FEL) model sparse update parameter splitting

来源：评论

学校读者我要写书评

暂无评论

FedRecI2C: A Novel Federated Recommendation Framework Integrating Communication and computation to Accelerate Convergence Under Communication Constraints

引用

FUTURE INTERNET 2025年第3期17卷 132-132页

作者： Zheng, Qizhong Huang, Xiujie Jinan Univ Coll Cyber Secur Guangzhou 511436 Peoples R China Jinan Univ Coll Informat Sci & Technol Guangzhou 510632 Peoples R China

The federated recommender system (FRS) employs federated learning methodologies to create a recommendation model in a distributed environment, where clients share locally updated data with the server without exposing raw data and achieving privacy preservation. However, varying communication capabilities among devices restrict the participation of only a subset of clients in each round of federated training, resulting in slower convergence and requiring additional training rounds. In this work, we propose a novel federated recommendation framework, called FedRecI2C, which integrates communication and computation resources in the system. This framework accelerates convergence by utilizing not only communication-capable clients for federated training but also communication-constrained clients to leverage their computation and limited communication resources for further local training. This framework offers simplicity and flexibility, providing a plug-and-play architecture that effectively enhances the convergence speed in FRSs. It has demonstrated remarkable effectiveness in a wide range of FRSs when operating under diverse communication conditions. Extensive experiments are conducted to validate the effectiveness of FedRecI2C. Moreover, we provide in-depth analyses of the FedRecI2C framework, offering novel insights into the training patterns of FRSs.

关键词： federated recommender system convergence rate communication resources computation resources

来源：评论

学校读者我要写书评

暂无评论

Q-learning based computation offloading for multi-UAV-enabled cloud-edge computing networks

引用

IET COMMUNICATIONS 2020年第15期14卷 2481-2490页

作者： Wang, Meng Shi, Shuo Gu, Shushi Gu, Xuemai Qin, Xue Harbin Inst Technol Sch Elect & Informat Engn Harbin 150001 Peoples R China Peng Cheng Lab Shenzhen 518055 Peoples R China Harbin Inst Technol Shenzen Sch Elect & Informat Engn Shenzhen 518055 Peoples R China Texas A&M Univ Sch Comp Sci Corpus Christi TX 78412 USA

Unmanned aerial vehicles (UAVs) have been recently considered as a flying platform to provide wide coverage and relaying services for mobile users (MUs). Mobile edge computing (MEC) is developed as a new paradigm to improve quality of experience of MUs in future networks. Motivated by the high flexibility and controllability of UAVs, in this study, the authors study a multi-UAV-enabled MEC system, in which UAVs have computation resources to offer computation offloading opportunities for MUs, aiming to reduce MUs' total consumptions in terms of time and energy. Considering the rich computation resource in the remote cloud centre, they propose the MUs-Edge-Cloud three-layer network architecture, where UAVs play the role of flying edge servers. Based on this framework, they formulate the computation offloading issue as a mixed-integer non-linear programming problem, which is difficult to obtain an optimal solution in general. To address this, they propose an efficientQ-learning based computation offloading algorithm (QCOA) to reduce the complexity of optimisation problem. Numerical results show that the proposed QCOA outperforms benchmark offloading policies (e.g. random offloading, traversal offloading). Furthermore, the proposed three-layer network architecture achieves a 5% benefits compared with the traditional two-layer network architecture in terms of MUs' energy and time consumptions.

关键词： cloud computing integer programming cellular radio nonlinear programming autonomous aerial vehicles mobile computing learning (artificial intelligence) relay networks (telecommunication) multi-robot systems optimisation problem complexity reduction QCOA quality of experience MUs-Edge-Cloud three-layer network architecture MU energy consumption two-layer network architecture traversal offloading random offloading efficient Q-learning based computation offloading algorithm mixed-integer nonlinear programming problem edge servers remote cloud centre computation resources multiUAV-enabled MEC system controllability mobile edge computing mobile users relaying services flying platform UAV unmanned aerial vehicles multiUAV-enabled cloud-edge computing networks

来源：评论

学校读者我要写书评

暂无评论

Efficient computation Resource Management in Mobile Edge-Cloud Computing

引用

IEEE INTERNET OF THINGS JOURNAL 2019年第2期6卷 3455-3466页

作者： Zhang, Yongmin Lan, Xiaolong Li, Yue Cai, Lin Pan, Jianping Univ Victoria Dept Elect & Comp Engn Victoria BC V8W 3P6 Canada Southwest Jiaotong Univ Sch Informat Sci & Technol Chengdu 611756 Sichuan Peoples R China Univ Victoria Dept Comp Sci Victoria BC V8W 3P6 Canada

We study the computation resource management problem in mobile edge-cloud computing networks. Mobile edge servers shall first satisfy the computation requirements of mobile users and Internet of Things (IoT) devices, and then wholesale redundant computation resources to the cloud networks to maximize their profit. Due to the coarse time granularity of wholesales, computation resource buyback may happen occasionally to deal with traffic bursts. Thus, the mobile edge servers need to make a tradeoff between the wholesale profit and the buyback cost. In this paper, the computation resource management problem is modeled as profit maximization. To solve this problem, we first analyze the relationship among the reserved computation resources, the computation tasks of mobile users and IoT devices, and the buyback cost. Then, we design an efficient wholesale scheme to determine the amount of the wholesaled computation resources, by which the total expected profit of the mobile edge server can be maximized. Given the reserved computation resources, we also propose a fast-convergent realtime buyback scheme for mobile edge servers to minimize the buyback cost. Finally, the simulation results show that our proposed efficient wholesale and buyback scheme can increase the total profit while guaranteeing the computation delay of all the computation tasks, especially when the computation workloads are time-varying.

关键词： Cloud networks computation resources Internet of Things (IoT) mobile edge server profit maximization

来源：评论

学校读者我要写书评

暂无评论

Energy-efficient BBU pool virtualisation for C-RAN with quality of service guarantees

引用

IET COMMUNICATIONS 2020年第1期14卷 11-20页

作者： Abdelhakam, Mostafa M. Elmesalawy, Mahmoud M. Elhattab, Mohamed K. Esmat, Haitham H. Helwan Univ Dept Elect Commun & Comp Engn Fac Engn Cairo Egypt Concordia Univ Dept Elect & Comp Engn Montreal PQ Canada

Cloud radio access network (C-RAN) has been introduced as a promising network paradigm for improving the spectral and energy efficiency of next-generation mobile systems. In C-RAN, the computation resources of the centralised baseband units (BBUs) can be virtualised and dynamically shared among cells for energy-efficient BBU pool utilisation. In this study, a BBU virtualisation scheme is proposed to minimise the total power consumption in the BBU pool subject to constraints on users' quality of service in terms of real-time requirements, individual fronthaul capacity and BBU capacity. As the BBU processing time and transmission delay for each user data can be compromised to meet the user's real-time requirements while minimising the BBU power consumption, a priori user association phase is proposed and formulated as an optimisation problem to maximise the users' transmission rate, and hence minimising their transmission delay. Then, the BBU processing allocation phase is formulated as a bin-packing problem to minimise the overall power consumption in the BBU pool. Since this problem is combinatorial, a heuristic algorithm is proposed based on best-fit-decreasing algorithm to solve it. Extensive simulations show that the proposed scheme outperforms the comparable ones in terms of power consumption with reduction up to 33%.

关键词： quality of service radio access networks resource allocation minimisation bin packing computational complexity cloud computing telecommunication power management telecommunication computing mobile computing virtualisation total power consumption BBU pool subject real-time requirements individual fronthaul capacity BBU capacity BBU processing time transmission delay user data BBU power consumption a priori user association phase BBU processing allocation phase energy-efficient BBU pool virtualisation service guarantees cloud radio access network C-RAN energy efficiency next-generation mobile systems computation resources centralised baseband units energy-efficient BBU pool utilisation BBU virtualisation scheme spectral efficiency network paradigm bin-packing problem best-fit-decreasing algorithm

来源：评论

学校读者我要写书评

暂无评论

Energy-efficiency fog computing resource allocation in cyber physical internet of things systems

引用

IET COMMUNICATIONS 2019年第13期13卷 2003-2011页

作者： Chen, Xincheng Zhou, Yuchen He, Bintao Lv, Lu Xi An Jiao Tong Univ Sch Elect & Informat Engn Xian Shaanxi Peoples R China Xidian Univ Sch Telecommun Engn Xian Shaanxi Peoples R China

Cyber physical internet of things systems (CPIoTs), taking advantages of cyber physical systems, have been considered as a promising technology to provide better interaction and interoperability among various machines. However, the development of CPIoTs suffers severely from big data. In this context, fog computing is proposed to handle the big data bottleneck of CPIoTs. In this study, the authors focus on the joint optimisation of the communication resources and computation resources in fog computing-based CPIoTs to maximise the overall system energy efficiency, in which multiple fog nodes and end users are taken into consideration. Moreover, since the channel estimation error will become serious with the expanding scale, the imperfect channel state information is considered in this study. The formulated optimisation problem is a mixed integer non-linear problem which is indeed non-deterministic polynomial hard, hence a probability distribution method is proposed to reformulate the problem into a non-probability form, and the resource allocation algorithm based on Dinkelbach algorithm and Lagrange duality approach is adopted to tackle the problem efficiently. The simulation results confirm the effectiveness of the proposed scheme, especially when the scales are enormous.

关键词： computational complexity nonlinear programming resource allocation channel estimation integer programming energy conservation open systems telecommunication power management cyber-physical systems Internet of Things statistical distributions multiple fog nodes channel estimation error imperfect channel state information formulated optimisation problem mixed integer nonlinear problem resource allocation algorithm cyber physical systems interaction interoperability big data bottleneck joint optimisation communication resources computation resources cyber physical Internet of Things systems fog computing-based CPIoT overall system energy efficiency nondeterministic polynomial hard probability distribution method nonprobability form Dinkelbach algorithm Lagrange duality approach

来源：评论

学校读者我要写书评

暂无评论

Hardware Mapping Strategies of PLC Programs in FPGAs

引用

IFAC-PapersOnLine 2018年第6期51卷 131-137页

作者： Milik, Adam Hrynkiewicz, Edward Institute of Electronics Silesian University of Technology of Gliwice Poland

The automatic implementation of fast operating PLCs with massive parallel hardware processing is expected by wide range of designers not familiar with details of computation algorithms design in hardware. The paper presents an approach to implement the fixed point arithmetic computations of control programs with ability of computation resource sharing. The method is based on developed intermediate graph representation for PLC programming languages. The graph representation of a PLC program allows to reveal and to implement parallel computations. Presented methods are dedicated for FPGA architectures and enables utilization of pipelined processing. It is based on multiple stages scheduling shown in detail. Presented methods are intended to improve the efficiency of computation resource utilization of the direct mapping method. The paper is concluded with experiments results that compare achieved performance and resource requirements. © 2018

关键词： Field programmable gate arrays (FPGA) Fixed point arithmetic Hardware Mapping Pipe linings Pipeline processing systems Programmable logic controllers Scheduling Automatic implementation computation algorithm computation resources Graph representation Parallel computation Pipelined processing Resource requirements Resource sharing

来源：评论

学校读者我要写书评

暂无评论

OS-Aware Automotive Controller Design Using Non-Uniform Sampling

引用

ACM TRANSACTIONS ON CYBER-PHYSICAL SYSTEMS 2018年第4期2卷 1–22页

作者： Chang, Wanli Goswami, Dip Chakraborty, Samarjit Hamann, Arne Singapore Inst Technol 10 Dover Dr Singapore 138683 Singapore Eindhoven Univ Technol POB 513 NL-5600 MB Eindhoven Netherlands Tech Univ Munich Arcisstr 21 D-80290 Munich Germany Robert Bosch GmbH D-70465 Stuttgart Germany

Automotive functionalities typically consist of a large set of periodic/cyclic tasks scheduled under a real-time operating system (OS). Many of the tasks are feedback control applications with stringent performance requirements. OSEK/VDX is a common class of automotive OS that offers preemptive periodic schedules supporting a pre-configured set of periods. The feedback controllers implemented onto such OSEK/VDX-compliant systems need to use one of the pre-configured (sampling) periods. A shorter period is often desired for a higher control performance, and this implies a higher processor load. For a given performance requirement, the longest sampling period that meets this requirement is the optimal one. Given a limited set of pre-configured periods, such optimal sampling periods are often not available, and the practice is to choose a shorter available period-leading to a higher processor load. To address this, we propose a controller that cyclically switches among the available periods, thereby leading to an average sampling period closer to the optimal one. This way, we reduce the processor load and are able to pack more control applications on the same processor. Themain challenge in this article is the design of such controllers that takes into account such cyclic switching of sampling periods (i.e., use non-uniform sampling). The controller needs to meet specified performance requirements (settling time) and system constraints (e.g., input saturation). Such a non-convex constrained controller optimization problem as raised in the OS-aware automotive systems design has not been addressed in the traditional optimal control literature. A novel approach based on adaptively parameterized particle swarm optimization (PSO) is proposed to solve it. Using the OS-aware controller design with non-uniform sampling, we show that a higher number of applications can be packed on a processor, which is of particular interest in the cost-sensitive automotive industry.

关键词： OSEK/VDX non-uniform sampling computation resources optimal control

来源：评论

学校读者我要写书评

暂无评论

An Architecture to use Easy Java-Javascript Simulations in New Devices

引用

IFAC-PapersOnLine 2015年第29期48卷 129-133页

作者： Saenz, Jacobo Esquembre, Francisco Garcia, Felix J. Torre, Luis De La Dormido, Sebastian Computer Science and Automatics Department Computer Science School UNED Juan del Rosal 16 Madrid Spain Mathematics Faculty Universidad de Murcia Campus de Espinardo Murcia30071 Spain Departament of Computer Engineering and Technology Informatics Faculty Universidad de Murcia Campus de Espinardo Murcia30071 Spain

Nowadays, virtual and remote laboratories play an important role in the students' learning process. For distance education in scientific areas, where the laboratories are an essential part, this importance is especially relevant. In the last years, the newly discovered Java vulnerabilities and the new devices which do not support Java applications (tablets and smart-phones), make even harder to run or develop these laboratories as Java applications. On the one hand, a non-digitally signed applet cannot be run in a web browser or in a smart-phone. On the other hand, portable devices, like a pc-tablet, have limited computation resources, and sometimes it is not enough to run complex simulations. EjsS (Easy Java-Javascript Simulations) is an open source tool used for creating all kind of simulations by introducing the equations of the model, and building a graphical user interface (GUI) for the application. This work presents a solution to the two previous problems that appear when using EjsS. Both are resolved using a Java model that runs in a server and takes the heavier computational load. This Java model is linked with a Javascript GUI in the client device. The link is based on a web-socket connection between server and client using a JSON format. The proposed solution has additional advantages, such as the possibility of multiple users working with the same model, or reusing an already existing Java application by just building a new Javascript view. © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.

关键词： Open source software Control engineering Digital computers Distance education Education Education computing Graphical user interfaces High level languages Laboratories Smartphones Telephone sets Complex simulation computation resources computational loads Graphical user interfaces (GUI) Open source tools Remote laboratories Simulation Socket connections

来源：评论

学校读者我要写书评

暂无评论

Smart Task Distributor for MapReduce on Cloud Computing

Smart Task Distributor for MapReduce on Cloud Computing

引用

International Conference on Cloud Computing and Big Data (CLOUDCOM-ASIA)

作者： Huang, Tzu-Chi Chu, Kuo-Chih Liang, Jun-Ming Lunghwa Univ Sci & Technol Dept Elect Engn Taoyuan Taoyuan County Taiwan

ISBN: (纸本)9781479928293

A MapReduce system is widely used to implement the large-scale computation on cloud computing. A MapReduce system currently defines computation resources in a node as a roughly configurable slot number, and distributes tasks over nodes according to the slot number. However, a MapReduce system may make computation resources of clusters in the underutilization or overutilization condition, because a task of different applications unlikely uses the same computation resources and because a node may have different CPUs with different capabilities. A MapReduce system can use Smart Task Distributor (STD) proposed in this paper to solve the problem of the computation resource underutilization or overutilization in clusters. Technically, a MapReduce system can use STD to smartly distribute tasks over nodes in clusters, because STD on the one hand gradually assigns tasks to a node in order to fully utilize computation resources in the node and on the other hand dynamically estimates the remaining computation resources in the node for toggling the assignment of tasks on demand without overloading it. In experiments, a MapReduce system is proved to get better performances with STD than with other ways.

关键词： STD computation resources Overutilization Underutilization MapReduce Cloud Computing

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：