-
公开(公告)号：US11734554B2
公开(公告)日：2023-08-22
申请号：US18047716
申请日：2022-10-19
发明人： Xiaoyu Yu , Yuwei Wang , Bo Zhang , Lixin Zhang
IPC分类号： G06N3/063 , G06N3/02 , G06F17/15 , G06F17/16 , G06F30/18 , G06F30/15 , G06F30/17 , G06F30/13 , G06F30/27
CPC分类号： G06N3/063 , G06F17/15 , G06F17/16 , G06F30/13 , G06F30/15 , G06F30/17 , G06F30/18 , G06F30/27 , G06N3/02
摘要： This application discloses a pooling processing method, applied to a pooling processing system of a convolutional neural network. The method includes receiving a feature map comprising a width dimension and a height dimension; establishing a data region comprising k+m columns and n rows, k, m, and n being integers; transferring data blocks each of size k in width and size n in height from the feature map in a width-height sequential scanning order into columns m+1 to k+m and rows 1 to n of the data region in pooling cycles; and in each pooling cycle corresponding to a data block transferred from the feature map to the data region: retaining columns k+1 to k+m of the data region of a previous pooling cycle to columns 1 to m of the data region; performing a first level pooling within each column of the data region to obtain k+m first level pooling values; and performing P second level pooling values by pooling P sets of first level pooling values selected according to a position of the data block relative to a width boundary in the feature map, P being an integer.
-
公开(公告)号：US11537857B2
公开(公告)日：2022-12-27
申请号：US16678726
申请日：2019-11-08
发明人： Xiaoyu Yu , Yuwei Wang , Bo Zhang , Lixin Zhang
IPC分类号： G06F30/18 , G06F30/13 , G06F30/15 , G06F30/17 , G06F30/27 , G06F17/15 , G06F17/16 , G06N3/063
摘要： This application discloses a pooling processing method, applied to a pooling processing system of a convolutional neural network. The pooling processing system includes a first storage device, a data region, a pooling computation kernel, and a pooling controller. The method includes: reading, by the pooling controller, k pieces of feature data from the first storage device in each reading cycle, the k pieces of feature data being components in a feature map generated by a convolution operation of the convolutional neural network, and k being an integer greater than 1; writing, by the pooling controller, the k pieces of feature data read from the first storage device into the data region, wherein the k pieces of feature data form one group among n groups of k pieces of data with each group arranged in a first dimension and the n groups arranged in a second dimension, wherein the n groups of k pieces of data are written into the data region in an updating cycle, wherein a duration of the updating cycle is n times a duration of the reading cycle, and wherein n cis an integer greater than 1; and transmitting, after the updating cycle is ended, data in the data region to the pooling computation kernel to perform a pooling operation, wherein the data in the data region comprises the n groups of k pieces of data and last m groups of data from a previous updating cycle with each group along the second dimension, wherein the last m groups of data are temporarily stored in the data region for use in pooling calculation by the pooling computation kernel in a next updating cycle. The technical solution in this application reduces the number of storage, numbers of reading and writing due to data reuses, and improves the efficiency of pooling processing.
-
公开(公告)号：US11537624B2
公开(公告)日：2022-12-27
申请号：US15988908
申请日：2018-05-24
发明人： Lixin Zhang , Leyu Lin , Feng Xia , Mu Yuan , Xiangcong Zeng , Zhe Feng
IPC分类号： G06F16/2457 , G06F16/335
摘要： Embodiments of the present invention disclose a result ranking method and device. The method includes: acquiring search key-information by using an interaction application, and searching a prestored interaction data set for at least one search result associated with the search key-information; ranking the at least one search result according to a quality assessment score corresponding to each of the at least one search result; and outputting the ranked at least one search result. The quality assessment score corresponding to each search result is a value generated according to a number of historical operations performed on each search result and an interactive influence score of an application identifier performing a historical operation on each search result.
-
公开(公告)号：US11507812B2
公开(公告)日：2022-11-22
申请号：US16885669
申请日：2020-05-28
发明人： Yu Meng , Yuwei Wang , Lixin Zhang , Xiaoyu Yu , Jianlin Gao , Jianping Zhu
摘要： The present disclosure describes methods, devices, and storage mediums for adjusting computing resource. The method includes obtaining an expected pooling time of a target pooling layer and a to-be-processed data volume of the target pooling layer; obtaining a current clock frequency corresponding to at least one computing resource unit used for pooling; determining a target clock frequency according to the expected pooling time of the target pooling layer and the to-be-processed data volume of the target pooling layer; and in response to that the convolution layer associated with the target pooling layer completes convolution and the current clock frequency is different from the target clock frequency, switching the current clock frequency of the at least one computing resource unit to the target clock frequency, and performing pooling in the target pooling layer based on the at least one computing resource unit having the target clock frequency.
-
公开(公告)号：US11200092B2
公开(公告)日：2021-12-14
申请号：US16888918
申请日：2020-06-01
发明人： Bo Zhang , Xiaoyu Yu , Yuwei Wang , Lixin Zhang
摘要： Embodiments of this application relate to a convolutional computing accelerator, a convolutional computing method, and a convolutional computing device, which belong to the technical field of electronic circuits. The convolutional computing accelerator includes: a controller, a computing matrix, and a first cache. The computing matrix comprising at least one row of computing units, each row of computing units comprising at least two adjacent connected computing units. The controller is configured to control input data of each row of computing units to be loaded into the first cache, and to control the input data loaded into the first cache to be inputted into the two adjacent computing units in a corresponding row. Each of the computing units in the corresponding row is configured to perform, in a first clock cycle, a convolutional computation based on received input data and a pre-stored convolutional kernel.
-
-
-
-
搜索结果
5 条记录 (耗时 0.016 秒)
