公开(公告)号：US20200050481A1

公开(公告)日：2020-02-13

申请号：US16506099

申请日：2019-07-09

Applicant: Beijing Baidu Netcom Science and Technology Co., Ltd.

Inventor： Jian Ouyang ,  Xueliang Du ,  Yingnan Xu ,  Huimin Li

IPC: G06F9/48 ,  G06F9/30 ,  G06F9/38 ,  G06F9/34

Abstract: Disclosed are a computing method applied to an artificial intelligence chip and the artificial intelligence chip. The method includes: a target processor core generating, in response to determining a computational identifier obtained by decoding a to-be-executed instruction being a preset complex computational identifier, a complex computational instruction using the computational identifier and at least one operand obtained by decoding, and adding the generated complex computational instruction to a complex computational instruction queue; and a computational accelerator selecting a complex computational instruction from the complex computational instruction queue, executing a complex computation indicated by the complex computational identifier in the selected complex computational instruction using the at least one operand in the selected complex computational instruction as an inputted parameter, to obtain computational result; and writing the obtained computational result as a complex computational result into a complex computational result queue.

公开(公告)号：US11210322B2

公开(公告)日：2021-12-28

申请号：US16814178

申请日：2020-03-10

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

Inventor： Huimin Li ,  Jian Ouyang

IPC: G06F17/30 ,  G06F16/28 ,  G06F16/22 ,  G06F16/901

Abstract: Embodiments of the present disclosure relate to a method and apparatus for reducing storage space of a parameter table. The method may include: storing the parameter table in a lookup table system configured to compute an output value of a non-linear function according to an input value of the non-linear function, the parameter table including only an index value associated with an input value on one side of a median in a domain of the non-linear function and a parameter value corresponding to the index value; determining, by using a corresponding relationship between the index value associated with the input value on one side and the parameter value corresponding to the index value, a parameter value corresponding to an index value associated with an input value on the other side; and computing the output value by using the input value on the other side and the determined corresponding parameter value.

公开(公告)号：US11651198B2

公开(公告)日：2023-05-16

申请号：US16502687

申请日：2019-07-03

Applicant: Beijing Baidu Netcom Science and Technology Co., Ltd.

Inventor： Jiaxin Shi ,  Huimin Li ,  Yong Wang

IPC: G06N3/048 ,  G06N3/08 ,  G06F7/483

CPC classification number: G06N3/048 ,  G06F7/483 ,  G06N3/08

Abstract: Embodiments of the present disclosure relate to a data processing method and apparatus for a neural network. The neural network is provided with at least one activation function. A method may include: converting, in response to that an activation function acquiring current data is a target function, based on a conversion relationship between the target function and a preset function, the current data into input data of the preset function; finding out first output data of the preset function with the input data as an input in a lookup table corresponding to the preset function; obtaining second output data of the target function with the current data as an input by conversion based on the conversion relationship and the first output data; and outputting the second output data.

公开(公告)号：US11221851B2

公开(公告)日：2022-01-11

申请号：US16936676

申请日：2020-07-23

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

Inventor： Huimin Li ,  Peng Wu ,  Jian Ouyang

IPC: G06F9/30 ,  G06F9/445 ,  G06F9/54

Abstract: Embodiments of the present disclosure provide a method, executed by a computing device, for configuring a vector operation, an apparatus, a device, and a storage medium. The method includes obtaining information indicating at least one configurable vector operation parameter. The information indicating the at least one configurable vector operation parameter indicates a type and a value of the configurable vector operation parameter. The method further includes: based on the type and the value of the configurable vector operation parameter, configuring multiple vector operation circuits to enable each of the vector operation circuits to execute a target vector operation including two or more basic vector operations and defined based on the type and value of the configurable vector operation parameter.

公开(公告)号：US20200050924A1

公开(公告)日：2020-02-13

申请号：US16502687

申请日：2019-07-03

Applicant: Beijing Baidu Netcom Science and Technology Co., Ltd.

Inventor： Jiaxin Shi ,  Huimin Li ,  Yong Wang

IPC: G06N3/04 ,  G06F7/483 ,  G06N3/08

Abstract: Embodiments of the present disclosure relate to a data processing method and apparatus for a neural network. The neural network is provided with at least one activation function. A method may include: converting, in response to that an activation function acquiring current data is a target function, based on a conversion relationship between the target function and a preset function, the current data into input data of the preset function; finding out first output data of the preset function with the input data as an input in a lookup table corresponding to the preset function; obtaining second output data of the target function with the current data as an input by conversion based on the conversion relationship and the first output data; and outputting the second output data.