Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
NEURAL NETWORK CALCULATION METHOD AND APPARATUS THEREFOR
Document Type and Number:
WIPO Patent Application WO/2022/173095
Kind Code:
A1
Abstract:
The present invention relates to a neural network calculation method performed by a neural network calculation apparatus, wherein the neural network calculation method comprises the steps of: separating input data in the depth direction and shifting the input data according to a pre-calculated gradient; and applying quantization to point-wise convolution of the shifted input data.

Inventors:
KIM EUN HUI (KR)
LEE KYONG HA (KR)
SUNG WON KYUNG (KR)
Application Number:
PCT/KR2021/016682
Publication Date:
August 18, 2022
Filing Date:
November 15, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
KOREA INST SCI & TECH INF (KR)
International Classes:
G06N3/063; G06N3/08
Domestic Patent References:
WO2020258527A12020-12-30
Foreign References:
KR20190087263A2019-07-24
US20190188237A12019-06-20
Other References:
WU BICHEN; WAN ALVIN; YUE XIANGYU; JIN PETER; ZHAO SICHENG; GOLMANT NOAH; GHOLAMINEJAD AMIR; GONZALEZ JOSEPH; KEUTZER KURT: "Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions", 2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, IEEE, 18 June 2018 (2018-06-18), pages 9127 - 9135, XP033473838, DOI: 10.1109/CVPR.2018.00951
TAO SHENG; CHEN FENG; SHAOJIE ZHUO; XIAOPENG ZHANG; LIANG SHEN; MICKEY ALEKSIC: "A Quantization-Friendly Separable Convolution for MobileNets", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 23 March 2018 (2018-03-23), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081409117, DOI: 10.1109/EMC2.2018.00011
KIM EUNHUI; LEE KYONG-HA: "Spatial Shift Point-Wise Quantization", IEEE ACCESS, IEEE, USA, vol. 8, 16 November 2020 (2020-11-16), USA , pages 207683 - 207690, XP011822322, DOI: 10.1109/ACCESS.2020.3038164
Attorney, Agent or Firm:
SHIM, Kyoung-Shik et al. (KR)
Download PDF: