US 11,868,428 B2
Apparatus and method with compressed neural network computation
Seon Min Rhee, Seoul (KR); Jaekyeom Kim, Seoul (KR); Gunhee Kim, Seoul (KR); Minjung Kim, Seoul (KR); Dongyeon Woo, Seoul (KR); and Seungju Han, Seoul (KR)
Assigned to Samsung Electronics Co., Ltd., Suwon-si (KR); and Seoul National University R&DB Foundation, Seoul (KR)
Filed by Samsung Electronics Co., Ltd., Suwon-si (KR); and Seoul National University R&DB Foundation, Seoul (KR)
Filed on Jun. 7, 2021, as Appl. No. 17/340,134.
Claims priority of application No. 10-2020-0090420 (KR), filed on Jul. 21, 2020; and application No. 10-2020-0166049 (KR), filed on Dec. 1, 2020.
Prior Publication US 2022/0027668 A1, Jan. 27, 2022
Int. Cl. G06V 10/40 (2022.01); G06F 18/211 (2023.01); G06N 3/08 (2023.01); G06F 21/34 (2013.01); G06F 18/22 (2023.01); G06F 18/214 (2023.01); G06V 10/74 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06V 40/16 (2022.01)
CPC G06F 18/211 (2023.01) [G06F 18/214 (2023.01); G06F 18/22 (2023.01); G06F 21/34 (2013.01); G06N 3/08 (2013.01); G06V 10/40 (2022.01); G06V 10/74 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06V 40/172 (2022.01)] 23 Claims
OG exemplary drawing
 
1. A processor-implemented method, the method comprising:
extracting feature data from input data using a first portion of a neural network;
generating compressed representation data of the extracted feature data by dropping a feature value from the extracted feature data at a drop layer of the neural network based on a drop probability corresponding to the feature value; and
indicating an inference result from the compressed representation data using a second portion of the neural network,
wherein the generating of the compressed representation data comprises determining whether to drop each feature value of the feature data based on a binomial distribution function with a drop probability corresponding to each feature value.