METHOD AND DEVICE FOR TRAINING MODEL OF QUASI-ALEXNET

WIPO Patent Application WO/2017/167114

A1

The embodiments of the invention provide a method and device for training a model of a quasi-Alexnet. The method comprises: utilizing a first graphical processing unit (GPU) to compute a first gradient and a second gradient in a quasi-Alexnet; receiving a third gradient in the quasi-Alexnet and transmitted by a second GPU; computing, according to the first gradient and the third gradient, a first model parameter of the quasi-Alexnet; receiving a fourth gradient in the quasi-Alexnet and transmitted by the second GPU; computing, according to the second gradient and the fourth gradient, a second model parameter of the quasi-Alexnet; utilizing the first model parameter and the second model parameter to train a model of the quasi-Alexnet to perform computation and communication processes separately. The embodiment can perform in parallel computation of a convolutional layer in the quasi-Alexnet and communication using fully connected layer parameters, effectively reducing the time consumed for training a model, and increasing operational efficiency of training the model.

WO/2022/019620	METHOD AND DEVICE FOR PROVIDING PERSONALIZED COOKING SERVICE
JP2020009415	SYSTEM AND METHOD OF IDENTIFYING MALICIOUS FILES USING LEARNING MODEL TRAINED ON MALICIOUS FILE
WO/2021/164481	NEURAL NETWORK MODEL-BASED AUTOMATIC HANDWRITTEN SIGNATURE VERIFICATION METHOD AND DEVICE

WANG SIYU (CN)

PCT/CN2017/077897

October 05, 2017

March 23, 2017

Click for automatic bibliography generation  

ALIBABA GROUP HOLDING LTD (CN)
WANG SIYU (CN)

G06N3/08

CN104036451A	2014-09-10
CN104035751A	2014-09-10
CN104463324A	2015-03-25
US20150161522A1	2015-06-11

CO-HORIZON INTELLECTUAL PROPERTY INC. (CN)

View/Download PDF      PDF Help

Previous Patent: TRANSPORTATION MEANS USING DEVICE FOR CONVERTING RECIPROCATING RECTILINEAR MOTION INTO SINGLE-DIRECT...

Next Patent: USER EVALUATION METHOD, DEVICE, AND APPARATUS