公开(公告)号：US12015728B2

公开(公告)日：2024-06-18

申请号：US17894582

申请日：2022-08-24

Applicant: Samsung Electronics Co., Ltd.

Inventor： Doyun Kim ,  Hokyung Kang ,  Dongwoo Lee ,  Jihyuk Jang

IPC: H04N23/50 ,  H04M1/02 ,  H04N23/51 ,  H04N23/54 ,  H04N23/55 ,  H04N23/67

CPC classification number: H04M1/0264 ,  H04N23/51 ,  H04N23/54 ,  H04N23/55 ,  H04N23/675

Abstract: An electronic device includes a camera module, wherein the camera module includes a lens assembly including lenses aligned along an optical axis, an actuator, a first magnet disposed on a first surface of the lens assembly, a first coil configured to move the lens assembly along the optical axis, a metal shield structure disposed on an outer surface of the actuator to face the first coil, and a fixing structure disposed on a third surface perpendicular to the first surface of the lens assembly, wherein at least a portion of the fixing structure has a magnetic property. When one end of the fixing structure is inserted into an opening of a housing, the fixing structure is fixed in an optical axis direction by the opening and is fixed in the first direction by a magnetic force acting on the metal shield structure.

公开(公告)号：US20210004663A1

公开(公告)日：2021-01-07

申请号：US16786462

申请日：2020-02-10

Applicant: Samsung Electronics Co., Ltd.

Inventor： Juhee Park ,  Doyun Kim ,  Sanghyuck Ha

IPC: G06N3/04 ,  G06N3/08 ,  G06K9/46 ,  G06N3/063

Abstract: A neural network device includes a quantization parameter calculator configured to quantize parameters of a neural network that is pre-trained, so that the quantized parameters are of mixed data types, analyze a statistical distribution of parameter values of an M-bit floating-point type, the parameter values being associated with at least one layer of the neural network, M being a natural number greater than three, obtain a quantization level of each of the parameters statistically covering a distribution range of the parameter values, based on the analyzed statistical distribution, and quantize input data and weights of the M-bit floating-point type into asymmetric input data of an N-bit fixed-point type and weights of an N-bit floating-point type, using quantization parameters that are obtained based on the obtained quantization level of each of the parameters, N being a natural number greater than one and less than M.

公开(公告)号：US11900262B2

公开(公告)日：2024-02-13

申请号：US16738038

申请日：2020-01-09

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Kyoungyoung Kim ,  Sangsoo Ko ,  Doyun Kim ,  Sanghyuck Ha

IPC: G06N3/10

CPC classification number: G06N3/10

Abstract: A neural network system for processing a neural network model including an operation processing graph that includes a plurality of operations, includes an operation processor including an internal memory storing a first module input feature map. The operation processor is configured to obtain a first branch output feature map by performing a first operation among the plurality of operations, based on the stored first module input feature map, and obtain a second branch output feature map by performing a second operation among the plurality of operations after the first operation is performed, based on the stored first module input feature map. The internal memory maintains storage of the first module input feature map while the first operation is performed.

公开(公告)号：US20200234147A1

公开(公告)日：2020-07-23

申请号：US16738038

申请日：2020-01-09

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Kyoungyoung Kim ,  Sangsoo Ko ,  Doyun Kim ,  Sanghyuck Ha

IPC: G06N3/10

Abstract: A neural network system for processing a neural network model including an operation processing graph that includes a plurality of operations, includes an operation processor including an internal memory storing a first module input feature map. The operation processor is configured to obtain a first branch output feature map by performing a first operation among the plurality of operations, based on the stored first module input feature map, and obtain a second branch output feature map by performing a second operation among the plurality of operations after the first operation is performed, based on the stored first module input feature map. The internal memory maintains storage of the first module input feature map while the first operation is performed.

公开(公告)号：US12073309B2

公开(公告)日：2024-08-27

申请号：US16786462

申请日：2020-02-10

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Juhee Park ,  Doyun Kim ,  Sanghyuck Ha

IPC: G06N3/063 ,  G06N3/047 ,  G06N3/08 ,  G06V10/70 ,  G06V10/82 ,  G06V10/44

CPC classification number: G06N3/063 ,  G06N3/047 ,  G06N3/08 ,  G06V10/70 ,  G06V10/82 ,  G06V10/454

Abstract: A neural network device includes a quantization parameter calculator configured to quantize parameters of a neural network that is pre-trained, so that the quantized parameters are of mixed data types, analyze a statistical distribution of parameter values of an M-bit floating-point type, the parameter values being associated with at least one layer of the neural network, M being a natural number greater than three, obtain a quantization level of each of the parameters statistically covering a distribution range of the parameter values, based on the analyzed statistical distribution, and quantize input data and weights of the M-bit floating-point type into asymmetric input data of an N-bit fixed-point type and weights of an N-bit floating-point type, using quantization parameters that are obtained based on the obtained quantization level of each of the parameters, N being a natural number greater than one and less than M.