On the Efficient Design of Neural Networks in Communication Systems

Weihong Xu; Xiaosi Tan; Yuxin Lin; Xiaohu You; Chuan Zhang; And Yair Be'Ery

doi:10.1109/IEEECONF44664.2019.9048768

On the Efficient Design of Neural Networks in Communication Systems

Weihong Xu
, Xiaosi Tan
, Yuxin Lin
, Xiaohu You
, Chuan Zhang
, And Yair Be'Ery

School of Electrical Engineering

Southeast University, Nanjing
Purple Mountain Laboratories

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

4 Scopus citations

Abstract

Various types of neural networks (NNs) have shown promising performance in communication systems. However, the low-latency implementation of these tasks is currently impractical due to the high computational complexity and large model size of NNs. In this paper, we propose an iterative optimization framework with retraining process to adaptively find the quantization scheme for different NNs. Moreover, the efficient design of convolutional neural networks is presented to reduce the required parameters and computational complexity. Experiment results for modulation classification, channel decoder and equalizer are presented. Compared to the original full-precision models, the quantized NN models achieve comparable performance with only 4 to 5 weight bits and 8-bit activation. The size of optimized models is significantly compressed and the hardware complexity of the NN inference is also reduced.

Original language	English
Title of host publication	Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019
Editors	Michael B. Matthews
Publisher	IEEE Computer Society
Pages	522-526
Number of pages	5
ISBN (Electronic)	9781728143002
DOIs	https://doi.org/10.1109/IEEECONF44664.2019.9048768
State	Published - Nov 2019
Event	53rd Asilomar Conference on Signals, Systems, and Computers, ACSSC 2019 - Pacific Grove, United States Duration: 3 Nov 2019 → 6 Nov 2019

Publication series

Name	Conference Record - Asilomar Conference on Signals, Systems and Computers
Volume	2019-November
ISSN (Electronic)	2576-2303

Conference

Conference	53rd Asilomar Conference on Signals, Systems, and Computers, ACSSC 2019
Country/Territory	United States
City	Pacific Grove
Period	3/11/19 → 6/11/19

Funding

Funders	Funder number
Jiangsu Provincial NSF	BK20180059
SRTP
National Natural Science Foundation of China	61871115, 61501116
Southeast University
Six Talent Peaks Project in Jiangsu Province	2018-DZXX-001
Fundamental Research Funds for the Central Universities

Keywords

Neural networks
communication
deep learning
depthwise separable convolution
quantization

Access to Document

10.1109/IEEECONF44664.2019.9048768

Cite this

Xu, W., Tan, X., Lin, Y., You, X., Zhang, C., & Be'Ery, A. Y. (2019). On the Efficient Design of Neural Networks in Communication Systems. In M. B. Matthews (Ed.), Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019 (pp. 522-526). Article 9048768 (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2019-November). IEEE Computer Society. https://doi.org/10.1109/IEEECONF44664.2019.9048768

@inproceedings{ae474ef3ed2a4699a55cb61f494cdce4,

title = "On the Efficient Design of Neural Networks in Communication Systems",

abstract = "Various types of neural networks (NNs) have shown promising performance in communication systems. However, the low-latency implementation of these tasks is currently impractical due to the high computational complexity and large model size of NNs. In this paper, we propose an iterative optimization framework with retraining process to adaptively find the quantization scheme for different NNs. Moreover, the efficient design of convolutional neural networks is presented to reduce the required parameters and computational complexity. Experiment results for modulation classification, channel decoder and equalizer are presented. Compared to the original full-precision models, the quantized NN models achieve comparable performance with only 4 to 5 weight bits and 8-bit activation. The size of optimized models is significantly compressed and the hardware complexity of the NN inference is also reduced.",

keywords = "Neural networks, communication, deep learning, depthwise separable convolution, quantization",

author = "Weihong Xu and Xiaosi Tan and Yuxin Lin and Xiaohu You and Chuan Zhang and Be'Ery, \{And Yair\}",

note = "Publisher Copyright: {\textcopyright} 2019 IEEE.; 53rd Asilomar Conference on Signals, Systems, and Computers, ACSSC 2019 ; Conference date: 03-11-2019 Through 06-11-2019",

year = "2019",

month = nov,

doi = "10.1109/IEEECONF44664.2019.9048768",

language = "אנגלית",

series = "Conference Record - Asilomar Conference on Signals, Systems and Computers",

publisher = "IEEE Computer Society",

pages = "522--526",

editor = "Matthews, \{Michael B.\}",

booktitle = "Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019",

address = "ארצות הברית",

}

Xu, W, Tan, X, Lin, Y, You, X, Zhang, C & Be'Ery, AY 2019, On the Efficient Design of Neural Networks in Communication Systems. in MB Matthews (ed.), Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019., 9048768, Conference Record - Asilomar Conference on Signals, Systems and Computers, vol. 2019-November, IEEE Computer Society, pp. 522-526, 53rd Asilomar Conference on Signals, Systems, and Computers, ACSSC 2019, Pacific Grove, United States, 3/11/19. https://doi.org/10.1109/IEEECONF44664.2019.9048768

On the Efficient Design of Neural Networks in Communication Systems. / Xu, Weihong; Tan, Xiaosi; Lin, Yuxin et al.
Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019. ed. / Michael B. Matthews. IEEE Computer Society, 2019. p. 522-526 9048768 (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2019-November).

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

TY - GEN

T1 - On the Efficient Design of Neural Networks in Communication Systems

AU - Xu, Weihong

AU - Tan, Xiaosi

AU - Lin, Yuxin

AU - You, Xiaohu

AU - Zhang, Chuan

AU - Be'Ery, And Yair

PY - 2019/11

Y1 - 2019/11

N2 - Various types of neural networks (NNs) have shown promising performance in communication systems. However, the low-latency implementation of these tasks is currently impractical due to the high computational complexity and large model size of NNs. In this paper, we propose an iterative optimization framework with retraining process to adaptively find the quantization scheme for different NNs. Moreover, the efficient design of convolutional neural networks is presented to reduce the required parameters and computational complexity. Experiment results for modulation classification, channel decoder and equalizer are presented. Compared to the original full-precision models, the quantized NN models achieve comparable performance with only 4 to 5 weight bits and 8-bit activation. The size of optimized models is significantly compressed and the hardware complexity of the NN inference is also reduced.

AB - Various types of neural networks (NNs) have shown promising performance in communication systems. However, the low-latency implementation of these tasks is currently impractical due to the high computational complexity and large model size of NNs. In this paper, we propose an iterative optimization framework with retraining process to adaptively find the quantization scheme for different NNs. Moreover, the efficient design of convolutional neural networks is presented to reduce the required parameters and computational complexity. Experiment results for modulation classification, channel decoder and equalizer are presented. Compared to the original full-precision models, the quantized NN models achieve comparable performance with only 4 to 5 weight bits and 8-bit activation. The size of optimized models is significantly compressed and the hardware complexity of the NN inference is also reduced.

KW - Neural networks

KW - communication

KW - deep learning

KW - depthwise separable convolution

KW - quantization

UR - https://www.scopus.com/pages/publications/85083309908

U2 - 10.1109/IEEECONF44664.2019.9048768

DO - 10.1109/IEEECONF44664.2019.9048768

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85083309908

T3 - Conference Record - Asilomar Conference on Signals, Systems and Computers

SP - 522

EP - 526

BT - Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019

A2 - Matthews, Michael B.

PB - IEEE Computer Society

T2 - 53rd Asilomar Conference on Signals, Systems, and Computers, ACSSC 2019

Y2 - 3 November 2019 through 6 November 2019

ER -

Xu W, Tan X, Lin Y, You X, Zhang C, Be'Ery AY. On the Efficient Design of Neural Networks in Communication Systems. In Matthews MB, editor, Conference Record of the 53rd Asilomar Conference on Signals, Systems and Computers, ACSSC 2019. IEEE Computer Society. 2019. p. 522-526. 9048768. (Conference Record - Asilomar Conference on Signals, Systems and Computers). doi: 10.1109/IEEECONF44664.2019.9048768

On the Efficient Design of Neural Networks in Communication Systems

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this