Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
METHOD AND DEVICE FOR TRAINING NEURAL NETWORK
Document Type and Number:
WIPO Patent Application WO/2018/228424
Kind Code:
A1
Abstract:
Provided is a method for training a neural network, which is applied to a resistive random access memory (RRAM), and reduces update operations during the training of the neural network by means of setting a pre-set threshold value, so as to extend the service life of the RRAM. The method comprises: inputting a neuron input value of an rth layer in the neural network to the RRAM, calculating, according to a filter in the RRAM, the neuron input value, to obtain a neuron output value of the rth layer in the neural network (301); performing calculation according to a kernel value of the RRAM, the neuron input value of the rth layer in the neural network, the neuron output value of the rth layer in the neural network and a back-propagation error value of the rth layer in the neural network, to obtain a back-propagation update value of the rth layer in the neural network (302); comparing the back-propagation update value of the rth layer in the neural network with the pre-set threshold value, and when the back-propagation update value of the rth layer in the neural network is greater than the pre-set threshold value (303), updating, according to the back-propagation update value of the rth layer in the neural network, the filter in the RRAM (304).

Inventors:
YAO JUN (CN)
LIU WULONG (CN)
WANG YU (CN)
XIA LIXUE (CN)
Application Number:
PCT/CN2018/091033
Publication Date:
December 20, 2018
Filing Date:
June 13, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
HUAWEI TECH CO LTD (CN)
International Classes:
G06N3/063; G06N3/08
Foreign References:
CN105303252A2016-02-03
CN106530210A2017-03-22
CN104376362A2015-02-25
US20120011089A12012-01-12
Other References:
See also references of EP 3627401A4
Download PDF: