公开(公告)号：US11861498B2

公开(公告)日：2024-01-02

申请号：US17968688

申请日：2022-10-18

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Guibin Wang ,  Shijun Cong ,  Hao Dong ,  Lei Jia

IPC: G06N3/08 ,  G06N3/04 ,  G06N3/045

CPC classification number: G06N3/08 ,  G06N3/045

Abstract: A method for compressing a neural network model includes acquiring a to-be-compressed neural network model. A first bit width, a second bit width and a target thinning rate corresponding to the to-be-compressed neural network model are determined. A target value is obtained according to the first bit width, the second bit width and the target thinning rate. Then the to-be-compressed neural network model is compressed using the target value, the first bit width and the second bit width to obtain a compression result of the to-be-compressed neural network model.

公开(公告)号：US20220188163A1

公开(公告)日：2022-06-16

申请号：US17653875

申请日：2022-03-08

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Shijun Cong ,  Guibin Wang ,  Xu Chen

IPC: G06F9/50

Abstract: A method for processing data includes: obtaining initial data; extracting data characteristics of the initial data; generating, based on the data characteristics, an initial memory consumption value required by a graphics processing unit (GPU) for processing the initial data; generating a recommendation processing parameter based on the initial memory consumption value; and obtaining target data by processing the initial data based on the recommendation processing parameter.

公开(公告)号：US20220207427A1

公开(公告)日：2022-06-30

申请号：US17655253

申请日：2022-03-17

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Yangkai Xu ,  Guibin Wang ,  Xiaoyin Fu ,  Zhijie Chen ,  Mingshun Yang ,  Shijun Cong ,  Ming Jia ,  Lei Jia

IPC: G06N20/00

Abstract: A method for training a data processing model includes: acquiring sample data; acquiring an initial data processing model, the initial data processing model including a plurality of forward nodes for outputting a plurality of intermediate results corresponding to the sample data; determining a plurality of time-dependent features corresponding to the plurality of forward nodes; acquiring a data processing model to be trained by processing the initial data processing model based on the plurality of time-dependent features; and training the data processing model to be trained using the sample data and the plurality of intermediate results.