中科微点-全球专利检索

  1. Login
  2. Sign Up

Search Results

237 records (took 0.032 seconds)

    Method and apparatus for accelerating distributed training of a deep neural network
    51.
    发明授权
    Method and apparatus for accelerating distributed training of a deep neural network 有权

    公开(公告)号：US11514309B2

    公开(公告)日：2022-11-29

    申请号：US16215033

    申请日：2018-12-10

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Jianxin Liao ,  Jingyu Wang ,  Jing Wang ,  Qi Qi ,  Jie Xu

    IPC: G06N3/08 ,  G06N3/04

    Abstract: Embodiments of the present invention provide a method and apparatus for accelerating distributed training of a deep neural network. The method comprises: based on parallel training, the training of deep neural network is designed as a distributed training mode. A deep neural network to be trained is divided into multiple sub-networks. A set of training samples is divided into multiple subsets of samples. The training of the deep neural network to be trained is performed with the multiple subsets of samples based on a distributed cluster architecture and a preset scheduling method. The multiple sub-networks are simultaneously trained so as to fulfill the distributed training of the deep neural network. The utilization of the distributed cluster architecture and the preset scheduling method may reduce, through data localization, the effect of network delay on the sub-networks under distributed training; adapt the training strategy in real time; and synchronize the sub-networks trained in parallel. As such, the time required for the distributed training of the deep neural network may be reduced and the training efficiency of the deep neural network may be improved.

    Method and Device of Network Resource Allocation
    52.
    发明申请
    Method and Device of Network Resource Allocation 有权

    公开(公告)号：US20220368650A1

    公开(公告)日：2022-11-17

    申请号：US17403479

    申请日：2021-08-16

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Hui YANG ,  Bowen BAO ,  Qiuyan YAO ,  Chao LI ,  Zhengjie SUN ,  Jie ZHANG

    IPC: H04L12/911 ,  H04L12/24 ,  G06N3/02

    Abstract: Disclosed is a method of network resource allocation. The method includes: generating an adjacency matrix of nodes in a metropolitan area network (MAN) according to spatial adjacency relationships of the nodes; generating a network state feature matrix according to traffic information of each node; extracting traffic spatial features of the nodes from the adjacency matrix and the network state feature matrix through a traffic spatial feature extraction model; obtaining predicted traffic of the nodes from the traffic spatial features through a traffic prediction model; and performing a network resource allocation according to the predicted traffic of the nodes. Further, a device of network resource allocation and a non-transitory computer-readable storage medium are also disclosed.

    DEEP REINFORCEMENT LEARNING-BASED INFORMATION PROCESSING METHOD AND APPARATUS FOR EDGE COMPUTING SERVER
    53.
    发明申请
    DEEP REINFORCEMENT LEARNING-BASED INFORMATION PROCESSING METHOD AND APPARATUS FOR EDGE COMPUTING SERVER 有权

    公开(公告)号：US20220255790A1

    公开(公告)日：2022-08-11

    申请号：US17520744

    申请日：2021-11-08

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Lanlan RUI ,  Yang YANG ,  Xuesong QIU ,  Jingyang YAN ,  Zhipeng GAO ,  Wenjing LI

    IPC: H04L12/24

    Abstract: A deep reinforcement learning-based information processing method includes: determining whether a target edge computing server enters an alert state according to a quantity of service requests received by the target edge computing server within a preset time period; when the target edge computing server enters the alert state, obtaining preset system status information from a preset memory library; computing an optimal action value corresponding to the target edge computing server based on a preset deep reinforcement learning model according to the preset system status information and preset strategy information; and generating an action corresponding to the target edge computing server according to the optimal action value, and performing the action on the target edge computing server. A deep reinforcement learning-based information processing apparatus for an edge computing server includes a first determining module, an acquisition module, a first computing module, a first generation module.

    Network resource scheduling method, apparatus, electronic device and storage medium
    54.
    发明授权
    Network resource scheduling method, apparatus, electronic device and storage medium 有权

    公开(公告)号：US11411865B2

    公开(公告)日：2022-08-09

    申请号：US16906867

    申请日：2020-06-19

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Jing Wang ,  Jingyu Wang ,  Haifeng Sun ,  Qi Qi ,  Bo He ,  Jianxin Liao

    IPC: H04L45/00 ,  H04L45/302

    Abstract: A network resource scheduling method, apparatus, an electronic device and a storage medium are disclosed. An embodiment of the method includes: upon receipt of a network data stream, determining a traffic type of the network data stream based on the number of data packets of the network data stream received within a specified period of time, lengths of the data packets and reception times of the data packets; for each data packet comprised in the network data stream, determining a target transmission path for the data packet, based on node state parameters of nodes in the network cluster, link state parameters of links in the network cluster, and the traffic type of the network data stream when the data packet is received; and transmitting the data packet via the target transmission path.

    Dynamic Production Scheduling Method and Apparatus Based on Deep Reinforcement Learning, and Electronic Device
    56.
    发明申请
    Dynamic Production Scheduling Method and Apparatus Based on Deep Reinforcement Learning, and Electronic Device 有权

    公开(公告)号：US20220179689A1

    公开(公告)日：2022-06-09

    申请号：US17524335

    申请日：2021-11-11

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Liang Liu ,  Xiaolong Zheng ,  Huadong Ma ,  Zihui Luo ,  Chengling Jiang

    IPC: G06F9/48 ,  G06N3/08

    Abstract: The embodiments of the present invention provide a dynamic production scheduling method, apparatus and electronic device based on deep reinforcement learning, which relate to the technical field of Industrial Internet of Things, and can reduce the overall processing time of jobs on the basis of not exceeding the processing capacity of production device. The embodiments of the present invention includes: acquiring static characteristics, dynamic characteristics of each of jobs and system dynamic characteristics, inputting the static characteristics, dynamic characteristics of each of jobs to be scheduled and system dynamic characteristics into a scheduling model to obtain a job execution sequence or batch execution sequence of the jobs in each production stage, wherein, the static characteristics of the job include an amount of tasks and time required for completion, the dynamic characteristics of the job include reception moment, and the system dynamic characteristics include a remaining amount of tasks that can be performed by the device in each production stage. The scheduling model is a model obtained after training a first actor network based on static characteristics and dynamic characteristics of a sample job, system dynamic characteristics, and a first critic network.

    Methods, apparatus and electronic devices for absolute time synchronization
    58.
    发明授权
    Methods, apparatus and electronic devices for absolute time synchronization 有权

    公开(公告)号：US11272471B2

    公开(公告)日：2022-03-08

    申请号：US17061358

    申请日：2020-10-01

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Xiangming Wen ,  Chenyu Zhang ,  Wei Zheng ,  Zhaoming Lu ,  Zhengying Wang ,  Cong Li

    IPC: H04W56/00 ,  H04J3/06

    Abstract: Embodiments of the present disclosure disclose a method, an apparatus, and an electronic device for an absolute time synchronization, the method comprising: receiving, from a baseband processing chip, a first timing signaling and a first real system frame information of the received first timing signaling, the first timing signaling comprising a first absolute time corresponding to local time of a base station at time of transmission of the first timing signaling, and a first reference system frame information used by the base station for transmitting the first timing signaling; calculating a first time delay between the base station and the terminal based on at least the first absolute time, if the first real system frame information is consistent with the first reference system frame information; receiving, from the baseband processing chip, subsequent timing signalings and corresponding second real system frame information used by the base station for transmitting the subsequent timing signalings to the terminal, the subsequent timing signalings each comprising a second absolution time corresponding to the local time of the base station at time of transmission of the subsequent timing signaling and a second reference system frame information used by the base station for transmitting the subsequent timing signaling; for each of the subsequent timing signalings, determining a time delay adjustment value between the subsequent timing signaling and the first timing signaling based on at least the first time delay and the second absolute time, if the second real time system frame information is consistent with the second reference system frame information; and after a number of the received subsequent timing signalings reaches a target number, obtaining an absolute time information based on the time delay adjustment values corresponding to the target number of subsequent timing signaling; wherein the absolute time information is used to synchronize time of the terminal to time of the base station.

    Clustering-based frequency deviation determination and elimination method and device, and electronic apparatus
    59.
    发明授权
    Clustering-based frequency deviation determination and elimination method and device, and electronic apparatus 有权

    公开(公告)号：US11218354B2

    公开(公告)日：2022-01-04

    申请号：US17270176

    申请日：2019-08-16

    Applicant: Beijing University of Posts and Telecommunications

    Inventor： Zhiyong Feng ,  Kezhong Zhang ,  Che Ji ,  Zhiqing Wei

    IPC: H04L5/00 ,  H04L27/26

    Abstract: Clustering-based methods, apparatuses for frequency offset determination and elimination and electronic devices are disclosed. The clustering-based method for frequency offset determination includes: determining a constellation diagram for a received signal; determining N different values within a preset frequency interval, as N frequency offset estimates for a frequency offset of the received signal; for each of the frequency offset estimates, correcting the constellation diagram based on the frequency offset estimate to obtain a corrected constellation diagram, clustering signal points on the corrected constellation diagram, and calculating an area of a signal region in the corrected constellation diagram after clustering; and determining a frequency offset estimate corresponding to a signal region with a minimum area as a value of the frequency offset. The embodiments of the present invention can improve the accuracy and stability of the calculated value of the frequency offset.

    Cross-layer network fault recovery system and method based on configuration migration
    60.
    发明申请
    Cross-layer network fault recovery system and method based on configuration migration 有权

    公开(公告)号：US20210281471A1

    公开(公告)日：2021-09-09

    申请号：US17192151

    申请日：2021-03-04

    Applicant: Beijing University of Posts and Telecommunications ,  State Grid Liaoning Power Co., Ltd. Dalian Power Supply Company

    Inventor： Rentao GU ,  Meng LIAN ,  Lin LIU ,  Jingzhao LUAN ,  Baoli WANG ,  Yuefeng JI

    IPC: H04L12/24 ,  H04L12/707 ,  H04L12/703 ,  H04L12/741 ,  H04L12/715

    Abstract: The invention discloses a cross-layer network fault recovery system and method based on configuration migration. The method includes: an upper layer switching network controller sends device fault information to a super controller; the super controller performs a fault device location, and selects a backup upper layer switching network device in an upper layer switching network; the super controller generates a third forwarding table, and delivers the third forwarding table to an upper layer switching network controller and an underlying layer switching network controller; the upper layer switching network controller generates a first forwarding table to be updated according to the third forwarding table, and delivers the first forwarding table to neighbor devices of the fault device and the backup upper layer switching network device; the underlying layer switching network controller calculates underlying layer switching network transmission channels according to the third forwarding table, and updates a second forwarding table of the underlying layer switching network devices on these transmission channels. The invention utilizes the reconfigurability of the underlying layer switching network to allow rapid replacement of the fault device and service recovery in the upper layer switching network.

Patent Agency Ranking