Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator

Baoting Li; Longjun Liu; Yanming Jin; Peng Gao; Hongbin Sun; Nanning Zheng

doi:10.1109/ASP-DAC47756.2020.9045739

Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator

Baoting Li
, Longjun Liu
, Yanming Jin
, Peng Gao
, Hongbin Sun
, Nanning Zheng

College of Artificial Intelligence and Robotics

Xi'an Jiaotong University

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

2 Scopus citations

Abstract

Network quantization is an effective solution to compress Deep Neural Networks (DNN) that can be accelerated with custom circuit. However, existing quantization methods suffer from significant loss in accuracy. In this paper, we propose an efficient shortcut architecture to enhance the representational capability of DNN between different convolution layers. We further implement the shortcut hardware architecture to effectively improve the accuracy of fully quantized neural networks accelerator. The experimental results show that our shortcut architecture can obviously improve network accuracy while increasing very few hardware resources ( 0.11 × and 0.17 × for LUT and FF respectively) compared with the whole accelerator.

Original language	English
Title of host publication	ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	289-294
Number of pages	6
ISBN (Electronic)	9781728141237
DOIs	https://doi.org/10.1109/ASP-DAC47756.2020.9045739
State	Published - Jan 2020
Event	25th Asia and South Pacific Design Automation Conference, ASP-DAC 2020 - Beijing, China Duration: 13 Jan 2020 → 16 Jan 2020

Publication series

Name	Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC
Volume	2020-January

Conference

Conference	25th Asia and South Pacific Design Automation Conference, ASP-DAC 2020
Country/Territory	China
City	Beijing
Period	13/01/20 → 16/01/20

Access to Document

10.1109/ASP-DAC47756.2020.9045739

Cite this

Li, B., Liu, L., Jin, Y., Gao, P., Sun, H., & Zheng, N. (2020). Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator. In ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings (pp. 289-294). Article 9045739 (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC; Vol. 2020-January). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ASP-DAC47756.2020.9045739

Li, Baoting ; Liu, Longjun ; Jin, Yanming et al. / Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator. ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings. Institute of Electrical and Electronics Engineers Inc., 2020. pp. 289-294 (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC).

@inproceedings{9d9918a22923472da1e9dd74eae36f05,

title = "Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator",

abstract = "Network quantization is an effective solution to compress Deep Neural Networks (DNN) that can be accelerated with custom circuit. However, existing quantization methods suffer from significant loss in accuracy. In this paper, we propose an efficient shortcut architecture to enhance the representational capability of DNN between different convolution layers. We further implement the shortcut hardware architecture to effectively improve the accuracy of fully quantized neural networks accelerator. The experimental results show that our shortcut architecture can obviously improve network accuracy while increasing very few hardware resources ( 0.11 × and 0.17 × for LUT and FF respectively) compared with the whole accelerator.",

author = "Baoting Li and Longjun Liu and Yanming Jin and Peng Gao and Hongbin Sun and Nanning Zheng",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 25th Asia and South Pacific Design Automation Conference, ASP-DAC 2020 ; Conference date: 13-01-2020 Through 16-01-2020",

year = "2020",

month = jan,

doi = "10.1109/ASP-DAC47756.2020.9045739",

language = "英语",

series = "Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "289--294",

booktitle = "ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings",

}

Li, B, Liu, L, Jin, Y, Gao, P, Sun, H & Zheng, N 2020, Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator. in ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings., 9045739, Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC, vol. 2020-January, Institute of Electrical and Electronics Engineers Inc., pp. 289-294, 25th Asia and South Pacific Design Automation Conference, ASP-DAC 2020, Beijing, China, 13/01/20. https://doi.org/10.1109/ASP-DAC47756.2020.9045739

Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator. / Li, Baoting; Liu, Longjun; Jin, Yanming et al.
ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings. Institute of Electrical and Electronics Engineers Inc., 2020. p. 289-294 9045739 (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC; Vol. 2020-January).

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

TY - GEN

T1 - Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator

AU - Li, Baoting

AU - Liu, Longjun

AU - Jin, Yanming

AU - Gao, Peng

AU - Sun, Hongbin

AU - Zheng, Nanning

PY - 2020/1

Y1 - 2020/1

N2 - Network quantization is an effective solution to compress Deep Neural Networks (DNN) that can be accelerated with custom circuit. However, existing quantization methods suffer from significant loss in accuracy. In this paper, we propose an efficient shortcut architecture to enhance the representational capability of DNN between different convolution layers. We further implement the shortcut hardware architecture to effectively improve the accuracy of fully quantized neural networks accelerator. The experimental results show that our shortcut architecture can obviously improve network accuracy while increasing very few hardware resources ( 0.11 × and 0.17 × for LUT and FF respectively) compared with the whole accelerator.

AB - Network quantization is an effective solution to compress Deep Neural Networks (DNN) that can be accelerated with custom circuit. However, existing quantization methods suffer from significant loss in accuracy. In this paper, we propose an efficient shortcut architecture to enhance the representational capability of DNN between different convolution layers. We further implement the shortcut hardware architecture to effectively improve the accuracy of fully quantized neural networks accelerator. The experimental results show that our shortcut architecture can obviously improve network accuracy while increasing very few hardware resources ( 0.11 × and 0.17 × for LUT and FF respectively) compared with the whole accelerator.

UR - https://www.scopus.com/pages/publications/85083035150

U2 - 10.1109/ASP-DAC47756.2020.9045739

DO - 10.1109/ASP-DAC47756.2020.9045739

M3 - 会议稿件

AN - SCOPUS:85083035150

T3 - Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC

SP - 289

EP - 294

BT - ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 25th Asia and South Pacific Design Automation Conference, ASP-DAC 2020

Y2 - 13 January 2020 through 16 January 2020

ER -

Li B, Liu L, Jin Y, Gao P, Sun H , Zheng N. Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator. In ASP-DAC 2020 - 25th Asia and South Pacific Design Automation Conference, Proceedings. Institute of Electrical and Electronics Engineers Inc. 2020. p. 289-294. 9045739. (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC). doi: 10.1109/ASP-DAC47756.2020.9045739

Designing Efficient Shortcut Architecture for Improving the Accuracy of Fully Quantized Neural Networks Accelerator

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this