公开(公告)号：US20210406646A1

公开(公告)日：2021-12-30

申请号：US17091338

申请日：2020-11-06

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Songyi HAN ,  Sangwon HA ,  Donghyuk KWON

IPC: G06N3/06 ,  G06N3/08 ,  G06K9/62 ,  G06F9/38

Abstract: A processor-implemented tensor processing method includes: receiving a request to process a neural network including a normalization layer by an accelerator; and generating an instruction executable by the accelerator in response to the request, wherein, by executing the instruction, the accelerator is configured to determine an intermediate tensor corresponding to a result of a portion of operations of the normalization layer, by performing, in a channel axis direction, a convolution based on an input tensor and a kernel, wherein the input tensor is of the normalization layer and includes a plurality of channels, a number of input channels of the kernel is determined based on the input tensor, and scaling values of elements of the kernel are determined based on the number of input channels.

公开(公告)号：US20210397935A1

公开(公告)日：2021-12-23

申请号：US17085264

申请日：2020-10-30

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN ,  Ihor VASYLTSOV

IPC: G06N3/063 ,  G06K9/62 ,  G06N5/04

Abstract: A processor-implemented tensor processing method includes: receiving a request to process a neural network including a normalization layer by an accelerator; and generating an instruction executable by the accelerator in response to the request, wherein, by executing the instruction, the accelerator is configured to determine an intermediate tensor corresponding to a result of performing a portion of operations included in the normalization layer, by performing, in a channel axis direction, a convolution based on: a target tensor on which the portion of operations is to be performed; and a kernel having a number of input channels and a number of output channels determined based on the target tensor and including elements of scaling values determined based on the target tensor.

公开(公告)号：US20210034971A1

公开(公告)日：2021-02-04

申请号：US16786006

申请日：2020-02-10

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN

IPC: G06N3/08 ,  G06N3/04

Abstract: A method of updating a neural network model by a terminal device, includes training a local model using a local data set collected by a terminal device to generate a trained local model; receiving, from a server, an independent identically distributed (i.i.d.) global data set, the i.i.d. global data set being a data set sampled for each class in a plurality of predefined classes; implementing the trained local model by inputting the i.i.d. global data set and transmitting final inference results of the implemented trained local model to the server; and receiving, from the server, a global model updated based on the final inference results of the inference.

公开(公告)号：US20240126833A1

公开(公告)日：2024-04-18

申请号：US18397716

申请日：2023-12-27

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN ,  Hyunsun PARK

IPC: G06F17/16 ,  G06F18/213 ,  G06N3/08

CPC classification number: G06F17/16 ,  G06F18/213 ,  G06N3/08

Abstract: A neural network apparatus for performing a matrix multiplication operation includes a memory having at least one program stored therein and a processor to perform one or more operations by executing the at least one program. The processor can determine whether to divide an initial weight in one of a column direction and a row direction according to whether a reshape operation and a transpose operation are performed before or after a matrix multiplication operation and generate division weights by dividing the initial weight by a head count in the determined direction. Also, the processor can generate intermediate feature maps by performing a matrix multiplication operation between the input feature map and the division weights and generate a final feature map based on the intermediate feature maps.

公开(公告)号：US20230274140A1

公开(公告)日：2023-08-31

申请号：US18144009

申请日：2023-05-05

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN ,  Seungwon LEE ,  Minkyoung CHO

IPC: G06N3/08 ,  G06N20/10

CPC classification number: G06N3/08 ,  G06N20/10

Abstract: A processor-implemented method of performing a convolution operation is provided. The method includes obtaining input feature map data and kernel data, determine the kernel data based on a number of input channels of the input feature map, a number of output channels of an output feature map, and a number of groups of the input feature map data and a number of groups of the kernel data related to the convolution operation, and performing the convolution operation based on the input feature map data and the determined kernel data.

公开(公告)号：US20210201132A1

公开(公告)日：2021-07-01

申请号：US16897461

申请日：2020-06-10

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN ,  Seungwon LEE ,  Minkyoung CHO

IPC: G06N3/08 ,  G06N20/10

Abstract: A processor-implemented method of performing a convolution operation is provided. The method includes obtaining input feature map data and kernel data, determine the kernel data based on a number of input channels of the input feature map, a number of output channels of an output feature map, and a number of groups of the input feature map data and a number of groups of the kernel data related to the convolution operation, and performing the convolution operation based on the input feature map data and the determined kernel data.

公开(公告)号：US20210173895A1

公开(公告)日：2021-06-10

申请号：US16851361

申请日：2020-04-17

Applicant: Samsung Electronics Co., Ltd.

Inventor： Songyi HAN ,  Hyunsun PARK

IPC: G06F17/16 ,  G06N3/08 ,  G06K9/62

Abstract: A neural network apparatus for performing a matrix multiplication operation includes a memory having at least one program stored therein and a processor to perform one or more operations by executing the at least one program. The processor can determine whether to divide an initial weight in one of a column direction and a row direction according to whether a reshape operation and a transpose operation are performed before or after a matrix multiplication operation and generate division weights by dividing the initial weight by a head count in the determined direction. Also, the processor can generate intermediate feature maps by performing a matrix multiplication operation between the input feature map and the division weights and generate a final feature map based on the intermediate feature maps.