Search International and National Patent Collections

1. (WO2017031630) DEEP CONVOLUTIONAL NEURAL NETWORK ACCELERATION AND COMPRESSION METHOD BASED ON PARAMETER QUANTIFICATION

Pub. No.:    WO/2017/031630    International Application No.:    PCT/CN2015/087792
Publication Date: Fri Mar 03 00:59:59 CET 2017 International Filing Date: Sat Aug 22 01:59:59 CEST 2015
IPC: G06T 7/00
Applicants: INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES
中国科学院自动化研究所
Inventors: CHENG, Jian
程健
WU, Jia Xiang
吴家祥
LENG, Cong
冷聪
LU, Han Qing
卢汉清
Title: DEEP CONVOLUTIONAL NEURAL NETWORK ACCELERATION AND COMPRESSION METHOD BASED ON PARAMETER QUANTIFICATION
Abstract:
The present invention provides a deep convolutional neural network acceleration and compression method based on parameter quantification, comprising: performing quantification on parameters of a deep convolutional neural network, to obtain multiple sub-codebooks and index values separately corresponding to the multiple sub-codebooks; and obtaining a feature graph of output of the deep convolutional neural network according to the multiple sub-codebooks and the index values separately corresponding to the multiple sub-codebooks. By means of the present invention, the acceleration and compression of a deep convolutional neural network can be implemented.