The parallelization of convolution on a CNN using a SIMT based GPGPU

This paper proposes a method to accelerate convolutional neural network(CNN) by utilizing GPGPU. The convolutional layer of the conventional CNN required a large number of multiplication operations. This paper seeks to reduce the number of multiplication operations through Winograd convolution opera...

Full description

Saved in:
Bibliographic Details
Published in2016 International SoC Design Conference (ISOCC) pp. 333 - 334
Main Authors Heekyeong Jeon, Kwanho Lee, Seonghyung Han, Kwangyeob Lee
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2016
Subjects
Acceleration
Computers
Convolution
convolutional neural network
gpu
Instruction sets
neural network
Neural networks
Parallel processing
parallelism
Signal processing algorithms
Online AccessGet full text

Cover

More Information
Summary:This paper proposes a method to accelerate convolutional neural network(CNN) by utilizing GPGPU. The convolutional layer of the conventional CNN required a large number of multiplication operations. This paper seeks to reduce the number of multiplication operations through Winograd convolution operation and perform parallel processing of the convolution operation by utilizing SIMT structure of GPGPU. The experiment was conducted using ModelSim and TestDrive, and the experimental results showed that the processing time was improved by about 7%, compared to the conventional convolution operation.
DOI:10.1109/ISOCC.2016.7799813