A federated learning (FL) system first needs to use heterogeneous clients to recruit data, and then perform data training locally on the client to finish model training for preserving users’ data privacy. Yet clients...
详细信息
A federated learning (FL) system first needs to use heterogeneous clients to recruit data, and then perform data training locally on the client to finish model training for preserving users’ data privacy. Yet clients still face the cost of computational energy consumption on model training and parameter transmission and are reluctant to join federated learning platform. Thus, incentive mechanisms are needed when inviting numerous clients to help train the global model. However, static pricing cannot adjust data size adaptively to minimize cost due to client random arrival for participation and incomplete information about computing cost. Moreover, how to balance data sampling time and model training for finite time horizons is under-explored. Therefore, in this paper, a two-phase FL model consisting of the data recruitment phase and model training phase is formulated to study the clients’ incentive mechanism design under incomplete information about clients’ arrivals and their private costs. First, we propose a dynamic pricing scheme for homogeneous clients that offer time-dependent monetary returns to clients by considering the tradeoff between the total payment to clients and the model accuracy loss. We show that the pricing should increase over time in the data-sampling phase due to data aging. Then we obtain the best time partition between data sampling and training phases under a closed-form dynamic pricing solution. It is shown that a small data recruitment threshold should be allocated if the client’ training time per global iteration is long. Further, the extension to heterogeneous clients with different data size and working time is discussed regarding to the optimal dynamic pricing and recruitment threshold. We prove monotonicity in multiple client-type choice and a linear algorithm is proposed to find the optimal recruitment threshold and client-type choice for total cost minimization. Finally, we present the robustness of our algorithm to data size and e
Within the context of cleaner production, enhancing energy efficiency and sustainability has emerged as a central focus in supercomputing center development. To address the challenges in predicting energy consumption,...
详细信息
GPUs have become the defacto hardware devices for accelerating Deep Neural Network (DNN) inference workloads. However, the conventional sequential execution mode of DNN operators in mainstream deep learning frameworks...
详细信息
WiFi Channel State Information (CSI)-based activity recognition has sparked numerous studies due to its widespread availability and privacy protection. However, when applied in practical applications, general CSI-base...
详细信息
With rising uncertainty in the real world, online Reinforcement Learning (RL) has been receiving increasing attention due to its fast learning capability and improving data efficiency. However, online RL often suffers...
详细信息
Time series is the main form of sensor data. Time series prediction is conducive to reduce the energy consumption of sensor nodes and increase the service life. As time cost and computation cost of the existing hybrid...
详细信息
ISBN:
(纸本)9781665405546
Time series is the main form of sensor data. Time series prediction is conducive to reduce the energy consumption of sensor nodes and increase the service life. As time cost and computation cost of the existing hybrid prediction methods for time series are not good, this paper proposes a hybrid prediction method based on MODWT-EMD for time series. Firstly, the original time series is decomposed into multiple frequency components by Maximal Overlapping Discrete Wavelet Transform (MODWT). Secondly, the multiple frequency components are decomposed successively by Empirical Mode Decomposition (EMD) to obtain IMF components and residuals. IMFs and residuals contain the feature information of time series. Then, the importance of the extracted feature information is scored and sorted by random forest. We select feature information with a higher score. Finally, the selected feature information is input into Bi-GRU for training. We predicted the original time series by the trained Bi-GRU and obtained the prediction results. The experimental results show that the method proposed in this paper can predict the time series effectively.
image captioning is usually based on the mainstream encoder-decoder framework, the accuracy of the encoders and the generation capability of the decoders directly affect the quality of image captions. But no matter ho...
详细信息
作者:
Liu, LinZhou, Jian-TaoXing, Hai-FengGuo, Xiao-YongCollege of Computer Science
Ecological Big Data Engineering Research Center of the Ministry of Education Cloud Computing and Service Software Engineering Laboratory of Inner Mongolia Autonomous Region National and Local Joint Engineering Research Center of Intelligent Information Processing Technology for Mongolian Social Computing and Data Processing Key Laboratory of Inner Mongolia Autonomous Region Big Data Analysis Technology Engineering Research Center of Inner Mongolia Autonomous Region Inner Mongolia University Inner Mongolia Hohhot China College of Computer Science and Technology
Inner Mongolia Normal University Inner Mongolia Hohhot China College of Computer Information and Management
Inner Mongolia University of Finance and Economics Inner Mongolia Hohhot China
Nowadays, the best-effort service can not guarantee the quality of service (QoS) for all kinds of services. QoS routing is an important method to guarantee QoS requirements. It involves path selection for flows based ...
详细信息
We discuss and analyze the virtual element method on general polygonal meshes for the time-dependent Poisson-Nernst-Planck equations, which are a nonlinear coupled system widely used in semiconductors and ion channels...
详细信息
Taking a military operation as an example, this paper makes an in-depth study of what tactics a single UAV should adopt when facing the interception of multiple enemy UAV groups. According to the analysis of the impac...
详细信息
ISBN:
(纸本)9781450387828
Taking a military operation as an example, this paper makes an in-depth study of what tactics a single UAV should adopt when facing the interception of multiple enemy UAV groups. According to the analysis of the impact diagram, the aerial range of the UAV standoff defines a threat area, that is, when the UAV arrives at the threat area, it needs to adopt a strategy for penetration, or it will be intercepted. The strategy adopted by UAV in the threat area is analyzed, and the decision control method is made.
暂无评论