Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems

Pei Ju Li; Lin Yi Jiang; Ting Yu Hu; Ding Chau Wang; Gwo Jiun Horng; Yu Chuan Lin; Min Hsiung Hung; Chao Chun Chen

doi:10.1007/978-981-19-8234-7_22

Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems

Pei Ju Li, Lin Yi Jiang, Ting Yu Hu, Ding Chau Wang, Gwo Jiun Horng, Yu Chuan Lin, Min Hsiung Hung, Chao Chun Chen

Institute of Manufacturing Information and Systems

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

We proposed two model compression methods by adjusting the structure of the deep network model, aiming to provide a small-sized model and realize the application requirement of edge computing. Edge Device has weak computing power and small storage capacity. The key idea of our proposed scheme is to minimize the computational cost of Expandableconvolutional-block neural network (ECB-Net) [7] to compress model size and optimize the model performance of ECB-Net. In order to achieve this, it is necessary to combine two model compression methods that adjust the ECB-Net network structure, including replacing Transpose Convolution in SIC-Block with SPP-Layer and replacing FC-Layer with TC-Layer. The experimental results reveal that the proposed scheme does have effective model compression capabilities and indeed prediction accuracy.

Original language	English
Title of host publication	Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings
Editors	Edward Szczerbicki, Krystian Wojtkiewicz, Sinh Van Nguyen, Marcin Pietranik, Marek Krótkiewicz
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	276-286
Number of pages	11
ISBN (Print)	9789811982330
DOIs	https://doi.org/10.1007/978-981-19-8234-7_22
Publication status	Published - 2022
Event	14th Asian Conference on Intelligent Information and Database Systems, ACIIDS 2022 - Ho Chi Minh City, Viet Nam Duration: 2022 Nov 28 → 2022 Nov 30

Publication series

Name	Communications in Computer and Information Science
Volume	1716 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	14th Asian Conference on Intelligent Information and Database Systems, ACIIDS 2022
Country/Territory	Viet Nam
City	Ho Chi Minh City
Period	22-11-28 → 22-11-30

All Science Journal Classification (ASJC) codes

General Computer Science
General Mathematics

Access to Document

10.1007/978-981-19-8234-7_22

Cite this

Li, P. J., Jiang, L. Y., Hu, T. Y., Wang, D. C., Horng, G. J., Lin, Y. C., Hung, M. H., & Chen, C. C. (2022). Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems. In E. Szczerbicki, K. Wojtkiewicz, S. V. Nguyen, M. Pietranik, & M. Krótkiewicz (Eds.), Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings (pp. 276-286). (Communications in Computer and Information Science; Vol. 1716 CCIS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-19-8234-7_22

Li, Pei Ju ; Jiang, Lin Yi ; Hu, Ting Yu et al. / Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems. Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings. editor / Edward Szczerbicki ; Krystian Wojtkiewicz ; Sinh Van Nguyen ; Marcin Pietranik ; Marek Krótkiewicz. Springer Science and Business Media Deutschland GmbH, 2022. pp. 276-286 (Communications in Computer and Information Science).

@inproceedings{246dd86767134520850f4a38c5adfe49,

title = "Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems",

abstract = "We proposed two model compression methods by adjusting the structure of the deep network model, aiming to provide a small-sized model and realize the application requirement of edge computing. Edge Device has weak computing power and small storage capacity. The key idea of our proposed scheme is to minimize the computational cost of Expandableconvolutional-block neural network (ECB-Net) [7] to compress model size and optimize the model performance of ECB-Net. In order to achieve this, it is necessary to combine two model compression methods that adjust the ECB-Net network structure, including replacing Transpose Convolution in SIC-Block with SPP-Layer and replacing FC-Layer with TC-Layer. The experimental results reveal that the proposed scheme does have effective model compression capabilities and indeed prediction accuracy.",

author = "Li, {Pei Ju} and Jiang, {Lin Yi} and Hu, {Ting Yu} and Wang, {Ding Chau} and Horng, {Gwo Jiun} and Lin, {Yu Chuan} and Hung, {Min Hsiung} and Chen, {Chao Chun}",

note = "Funding Information: This work was supported by Ministry of Science and Technology (MOST) of Taiwan under Grants 111-2221-E-006-202, 110-2221-E-034-009-MY2. This work was financially supported by the “Intelligent Manufacturing Research Center” (iMRC) in NCKU from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education in Taiwan. Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.; 14th Asian Conference on Intelligent Information and Database Systems, ACIIDS 2022 ; Conference date: 28-11-2022 Through 30-11-2022",

year = "2022",

doi = "10.1007/978-981-19-8234-7_22",

language = "English",

isbn = "9789811982330",

series = "Communications in Computer and Information Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "276--286",

editor = "Edward Szczerbicki and Krystian Wojtkiewicz and Nguyen, {Sinh Van} and Marcin Pietranik and Marek Kr{\'o}tkiewicz",

booktitle = "Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings",

address = "Germany",

}

Li, PJ, Jiang, LY, Hu, TY, Wang, DC, Horng, GJ, Lin, YC, Hung, MH & Chen, CC 2022, Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems. in E Szczerbicki, K Wojtkiewicz, SV Nguyen, M Pietranik & M Krótkiewicz (eds), Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings. Communications in Computer and Information Science, vol. 1716 CCIS, Springer Science and Business Media Deutschland GmbH, pp. 276-286, 14th Asian Conference on Intelligent Information and Database Systems, ACIIDS 2022, Ho Chi Minh City, Viet Nam, 22-11-28. https://doi.org/10.1007/978-981-19-8234-7_22

Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems. / Li, Pei Ju; Jiang, Lin Yi; Hu, Ting Yu et al.
Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings. ed. / Edward Szczerbicki; Krystian Wojtkiewicz; Sinh Van Nguyen; Marcin Pietranik; Marek Krótkiewicz. Springer Science and Business Media Deutschland GmbH, 2022. p. 276-286 (Communications in Computer and Information Science; Vol. 1716 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems

AU - Li, Pei Ju

AU - Jiang, Lin Yi

AU - Hu, Ting Yu

AU - Wang, Ding Chau

AU - Horng, Gwo Jiun

AU - Lin, Yu Chuan

AU - Hung, Min Hsiung

AU - Chen, Chao Chun

N1 - Funding Information: This work was supported by Ministry of Science and Technology (MOST) of Taiwan under Grants 111-2221-E-006-202, 110-2221-E-034-009-MY2. This work was financially supported by the “Intelligent Manufacturing Research Center” (iMRC) in NCKU from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education in Taiwan. Publisher Copyright: © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

PY - 2022

Y1 - 2022

N2 - We proposed two model compression methods by adjusting the structure of the deep network model, aiming to provide a small-sized model and realize the application requirement of edge computing. Edge Device has weak computing power and small storage capacity. The key idea of our proposed scheme is to minimize the computational cost of Expandableconvolutional-block neural network (ECB-Net) [7] to compress model size and optimize the model performance of ECB-Net. In order to achieve this, it is necessary to combine two model compression methods that adjust the ECB-Net network structure, including replacing Transpose Convolution in SIC-Block with SPP-Layer and replacing FC-Layer with TC-Layer. The experimental results reveal that the proposed scheme does have effective model compression capabilities and indeed prediction accuracy.

AB - We proposed two model compression methods by adjusting the structure of the deep network model, aiming to provide a small-sized model and realize the application requirement of edge computing. Edge Device has weak computing power and small storage capacity. The key idea of our proposed scheme is to minimize the computational cost of Expandableconvolutional-block neural network (ECB-Net) [7] to compress model size and optimize the model performance of ECB-Net. In order to achieve this, it is necessary to combine two model compression methods that adjust the ECB-Net network structure, including replacing Transpose Convolution in SIC-Block with SPP-Layer and replacing FC-Layer with TC-Layer. The experimental results reveal that the proposed scheme does have effective model compression capabilities and indeed prediction accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85144228760&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85144228760&partnerID=8YFLogxK

U2 - 10.1007/978-981-19-8234-7_22

DO - 10.1007/978-981-19-8234-7_22

M3 - Conference contribution

AN - SCOPUS:85144228760

SN - 9789811982330

T3 - Communications in Computer and Information Science

SP - 276

EP - 286

BT - Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings

A2 - Szczerbicki, Edward

A2 - Wojtkiewicz, Krystian

A2 - Nguyen, Sinh Van

A2 - Pietranik, Marcin

A2 - Krótkiewicz, Marek

PB - Springer Science and Business Media Deutschland GmbH

T2 - 14th Asian Conference on Intelligent Information and Database Systems, ACIIDS 2022

Y2 - 28 November 2022 through 30 November 2022

ER -

Li PJ, Jiang LY, Hu TY, Wang DC, Horng GJ, Lin YC et al. Expandable-Convolutional-Block Neural Inference with Spatial-Fused Multi-resolution Features and Terse Convolution for IoT-Based Intelligent Automated Systems. In Szczerbicki E, Wojtkiewicz K, Nguyen SV, Pietranik M, Krótkiewicz M, editors, Recent Challenges in Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings. Springer Science and Business Media Deutschland GmbH. 2022. p. 276-286. (Communications in Computer and Information Science). doi: 10.1007/978-981-19-8234-7_22