CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks
The explosive computation and memory requirements of convolutional neural networks (CNNs) hinder their deployment in resource-constrained devices. Because conventional CNNs perform identical parallelized computations even on redundant pixels, the saliency of various features in an image should be re...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2022-08-01
|
Series: | Electronics |
Subjects: | |
Online Access: | https://www.mdpi.com/2079-9292/11/17/2678 |
_version_ | 1797495896686985216 |
---|---|
author | Sangmin Hyun Chang Ho Ryu Ju Yeon Kang Hyun Jo Lim Tae Hee Han |
author_facet | Sangmin Hyun Chang Ho Ryu Ju Yeon Kang Hyun Jo Lim Tae Hee Han |
author_sort | Sangmin Hyun |
collection | DOAJ |
description | The explosive computation and memory requirements of convolutional neural networks (CNNs) hinder their deployment in resource-constrained devices. Because conventional CNNs perform identical parallelized computations even on redundant pixels, the saliency of various features in an image should be reflected for higher energy efficiency and market penetration. This paper proposes a novel channel and spatial gating network (CSGN) for adaptively selecting vital channels and generating spatial-wise execution masks. A CSGN can be characterized as a dynamic channel and a spatial-aware gating module by maximally utilizing opportunistic sparsity. Extensive experiments were conducted on the CIFAR-10 and ImageNet datasets based on ResNet. The results revealed that, with the proposed architecture, the amount of multiply-accumulate (MAC) operations was reduced by 1.97–11.78× and 1.37–13.12× on CIFAR-10 and ImageNet, respectively, with negligible accuracy degradation in the inference stage compared with the baseline architectures. |
first_indexed | 2024-03-10T01:56:09Z |
format | Article |
id | doaj.art-a1b44add556047638d7f034807ca1091 |
institution | Directory Open Access Journal |
issn | 2079-9292 |
language | English |
last_indexed | 2024-03-10T01:56:09Z |
publishDate | 2022-08-01 |
publisher | MDPI AG |
record_format | Article |
series | Electronics |
spelling | doaj.art-a1b44add556047638d7f034807ca10912023-11-23T12:57:26ZengMDPI AGElectronics2079-92922022-08-011117267810.3390/electronics11172678CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural NetworksSangmin Hyun0Chang Ho Ryu1Ju Yeon Kang2Hyun Jo Lim3Tae Hee Han4Department of Artificial Intelligence, Sungkyunkwan University, Suwon 16419, KoreaDepartment of Artificial Intelligence, Sungkyunkwan University, Suwon 16419, KoreaDepartment of Electrical and Computer Engineering, Sungkyunkwan University, Suwon 16419, KoreaDepartment of Artificial Intelligence, Sungkyunkwan University, Suwon 16419, KoreaDepartment of Semiconductor Systems Engineering, Sungkyunkwan University, Suwon 16419, KoreaThe explosive computation and memory requirements of convolutional neural networks (CNNs) hinder their deployment in resource-constrained devices. Because conventional CNNs perform identical parallelized computations even on redundant pixels, the saliency of various features in an image should be reflected for higher energy efficiency and market penetration. This paper proposes a novel channel and spatial gating network (CSGN) for adaptively selecting vital channels and generating spatial-wise execution masks. A CSGN can be characterized as a dynamic channel and a spatial-aware gating module by maximally utilizing opportunistic sparsity. Extensive experiments were conducted on the CIFAR-10 and ImageNet datasets based on ResNet. The results revealed that, with the proposed architecture, the amount of multiply-accumulate (MAC) operations was reduced by 1.97–11.78× and 1.37–13.12× on CIFAR-10 and ImageNet, respectively, with negligible accuracy degradation in the inference stage compared with the baseline architectures.https://www.mdpi.com/2079-9292/11/17/2678dynamic neural networkexploiting sparsityefficient computationgating architecture |
spellingShingle | Sangmin Hyun Chang Ho Ryu Ju Yeon Kang Hyun Jo Lim Tae Hee Han CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks Electronics dynamic neural network exploiting sparsity efficient computation gating architecture |
title | CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks |
title_full | CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks |
title_fullStr | CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks |
title_full_unstemmed | CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks |
title_short | CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks |
title_sort | csgn combined channel and spatial wise dynamic gating architecture for convolutional neural networks |
topic | dynamic neural network exploiting sparsity efficient computation gating architecture |
url | https://www.mdpi.com/2079-9292/11/17/2678 |
work_keys_str_mv | AT sangminhyun csgncombinedchannelandspatialwisedynamicgatingarchitectureforconvolutionalneuralnetworks AT changhoryu csgncombinedchannelandspatialwisedynamicgatingarchitectureforconvolutionalneuralnetworks AT juyeonkang csgncombinedchannelandspatialwisedynamicgatingarchitectureforconvolutionalneuralnetworks AT hyunjolim csgncombinedchannelandspatialwisedynamicgatingarchitectureforconvolutionalneuralnetworks AT taeheehan csgncombinedchannelandspatialwisedynamicgatingarchitectureforconvolutionalneuralnetworks |