SDChannelNets: Extremely Small and Efficient Convolutional Neural Networks

JianNan ZHANG; JiJun ZHOU; JianFeng WU; ShengYing YANG

doi:10.1587/transinf.2019EDL8120

SDChannelNets: Extremely Small and Efficient Convolutional Neural Networks

JianNan ZHANG, JiJun ZHOU, JianFeng WU, ShengYing YANG

Full Text Views

0

Cite this

Summary :

Convolutional neural networks (CNNS) have a strong ability to understand and judge images. However, the enormous parameters and computation of CNNS have limited its application in resource-limited devices. In this letter, we used the idea of parameter sharing and dense connection to compress the parameters in the convolution kernel channel direction, thus greatly reducing the number of model parameters. On this basis, we designed Shared and Dense Channel-wise Convolutional Networks (SDChannelNets), mainly composed of Depth-wise Separable SD-Channel-wise Convolution layer. The advantage of SDChannelNets is that the number of model parameters is greatly reduced without or with little loss of accuracy. We also introduced a hyperparameter that can effectively balance the number of parameters and the accuracy of a model. We evaluated the model proposed by us through two popular image recognition tasks (CIFAR-10 and CIFAR-100). The results showed that SDChannelNets had similar accuracy to other CNNs, but the number of parameters was greatly reduced.

Publication: IEICE TRANSACTIONS on Information Vol.E102-D No.12 pp.2646-2650

Publication Date: 2019/12/01

Publicized: 2019/09/10

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2019EDL8120

Type of Manuscript: LETTER

Category: Biocybernetics, Neurocomputing

Authors

JianNan ZHANG
  Hangzhou Dianzi University
JiJun ZHOU
  Hangzhou Dianzi University
JianFeng WU
  Hangzhou Dianzi University
ShengYing YANG
  Hangzhou Dianzi University

Keyword

convolutional neural networks, parameter sharing, convolution kernel, reducing the number of model parameters

Cite this

Copy

JianNan ZHANG, JiJun ZHOU, JianFeng WU, ShengYing YANG, "SDChannelNets: Extremely Small and Efficient Convolutional Neural Networks" in IEICE TRANSACTIONS on Information, vol. E102-D, no. 12, pp. 2646-2650, December 2019, doi: 10.1587/transinf.2019EDL8120.
Abstract: Convolutional neural networks (CNNS) have a strong ability to understand and judge images. However, the enormous parameters and computation of CNNS have limited its application in resource-limited devices. In this letter, we used the idea of parameter sharing and dense connection to compress the parameters in the convolution kernel channel direction, thus greatly reducing the number of model parameters. On this basis, we designed Shared and Dense Channel-wise Convolutional Networks (SDChannelNets), mainly composed of Depth-wise Separable SD-Channel-wise Convolution layer. The advantage of SDChannelNets is that the number of model parameters is greatly reduced without or with little loss of accuracy. We also introduced a hyperparameter that can effectively balance the number of parameters and the accuracy of a model. We evaluated the model proposed by us through two popular image recognition tasks (CIFAR-10 and CIFAR-100). The results showed that SDChannelNets had similar accuracy to other CNNs, but the number of parameters was greatly reduced.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDL8120/_p

Copy

@ARTICLE{e102-d_12_2646,
author={JianNan ZHANG, JiJun ZHOU, JianFeng WU, ShengYing YANG, },
journal={IEICE TRANSACTIONS on Information},
title={SDChannelNets: Extremely Small and Efficient Convolutional Neural Networks},
year={2019},
volume={E102-D},
number={12},
pages={2646-2650},
abstract={Convolutional neural networks (CNNS) have a strong ability to understand and judge images. However, the enormous parameters and computation of CNNS have limited its application in resource-limited devices. In this letter, we used the idea of parameter sharing and dense connection to compress the parameters in the convolution kernel channel direction, thus greatly reducing the number of model parameters. On this basis, we designed Shared and Dense Channel-wise Convolutional Networks (SDChannelNets), mainly composed of Depth-wise Separable SD-Channel-wise Convolution layer. The advantage of SDChannelNets is that the number of model parameters is greatly reduced without or with little loss of accuracy. We also introduced a hyperparameter that can effectively balance the number of parameters and the accuracy of a model. We evaluated the model proposed by us through two popular image recognition tasks (CIFAR-10 and CIFAR-100). The results showed that SDChannelNets had similar accuracy to other CNNs, but the number of parameters was greatly reduced.},
keywords={},
doi={10.1587/transinf.2019EDL8120},
ISSN={1745-1361},
month={December},}

Copy

TY - JOUR
TI - SDChannelNets: Extremely Small and Efficient Convolutional Neural Networks
T2 - IEICE TRANSACTIONS on Information
SP - 2646
EP - 2650
AU - JianNan ZHANG
AU - JiJun ZHOU
AU - JianFeng WU
AU - ShengYing YANG
PY - 2019
DO - 10.1587/transinf.2019EDL8120
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E102-D
IS - 12
JA - IEICE TRANSACTIONS on Information
Y1 - December 2019
AB - Convolutional neural networks (CNNS) have a strong ability to understand and judge images. However, the enormous parameters and computation of CNNS have limited its application in resource-limited devices. In this letter, we used the idea of parameter sharing and dense connection to compress the parameters in the convolution kernel channel direction, thus greatly reducing the number of model parameters. On this basis, we designed Shared and Dense Channel-wise Convolutional Networks (SDChannelNets), mainly composed of Depth-wise Separable SD-Channel-wise Convolution layer. The advantage of SDChannelNets is that the number of model parameters is greatly reduced without or with little loss of accuracy. We also introduced a hyperparameter that can effectively balance the number of parameters and the accuracy of a model. We evaluated the model proposed by us through two popular image recognition tasks (CIFAR-10 and CIFAR-100). The results showed that SDChannelNets had similar accuracy to other CNNs, but the number of parameters was greatly reduced.
ER -