Fast implementation of neural network classification

Guiwon Seo; Jiheon Ok; Chulhee Lee

doi:10.1117/12.2026666

Fast implementation of neural network classification

Guiwon Seo, Jiheon Ok, Chulhee Lee

Department of Electrical and Electronic Engineering

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

Abstract

Most artificial neural networks use a nonlinear activation function that includes sigmoid and hyperbolic tangent functions. Most artificial networks employ nonlinear functions such as these sigmoid and hyperbolic tangent functions, which incur high complexity costs, particularly during hardware implementation. In this paper, we propose new polynomial approximation methods for nonlinear activation functions that can substantially reduce complexity without sacrificing performance. The proposed approximation methods were applied to pattern classification problems. Experimental results show that the processing time was reduced by up to 50% without any performance degradations in terms of computer simulation.

Original language	English
Title of host publication	Satellite Data Compression, Communications, and Processing IX
Publisher	SPIE
ISBN (Print)	9780819497215
DOIs	https://doi.org/10.1117/12.2026666
Publication status	Published - 2013
Event	Satellite Data Compression, Communications, and Processing IX - San Diego, CA, United States Duration: 2013 Aug 26 → 2013 Aug 27

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	8871
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Other

Other	Satellite Data Compression, Communications, and Processing IX
Country	United States
City	San Diego, CA
Period	13/8/26 → 13/8/27

All Science Journal Classification (ASJC) codes

Electronic, Optical and Magnetic Materials
Condensed Matter Physics
Computer Science Applications
Applied Mathematics
Electrical and Electronic Engineering

Access to Document

10.1117/12.2026666

Fingerprint Dive into the research topics of 'Fast implementation of neural network classification'. Together they form a unique fingerprint.

Cite this

@inproceedings{a6b3dc2f71e948efa77b0bcb8b75c923,

title = "Fast implementation of neural network classification",

abstract = "Most artificial neural networks use a nonlinear activation function that includes sigmoid and hyperbolic tangent functions. Most artificial networks employ nonlinear functions such as these sigmoid and hyperbolic tangent functions, which incur high complexity costs, particularly during hardware implementation. In this paper, we propose new polynomial approximation methods for nonlinear activation functions that can substantially reduce complexity without sacrificing performance. The proposed approximation methods were applied to pattern classification problems. Experimental results show that the processing time was reduced by up to 50% without any performance degradations in terms of computer simulation.",

author = "Guiwon Seo and Jiheon Ok and Chulhee Lee",

year = "2013",

doi = "10.1117/12.2026666",

language = "English",

isbn = "9780819497215",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

booktitle = "Satellite Data Compression, Communications, and Processing IX",

address = "United States",

}

Seo, G, Ok, J & Lee, C 2013, Fast implementation of neural network classification. in Satellite Data Compression, Communications, and Processing IX., 887107, Proceedings of SPIE - The International Society for Optical Engineering, vol. 8871, SPIE, Satellite Data Compression, Communications, and Processing IX, San Diego, CA, United States, 13/8/26. https://doi.org/10.1117/12.2026666

TY - GEN

T1 - Fast implementation of neural network classification

AU - Seo, Guiwon

AU - Ok, Jiheon

AU - Lee, Chulhee

PY - 2013

Y1 - 2013

N2 - Most artificial neural networks use a nonlinear activation function that includes sigmoid and hyperbolic tangent functions. Most artificial networks employ nonlinear functions such as these sigmoid and hyperbolic tangent functions, which incur high complexity costs, particularly during hardware implementation. In this paper, we propose new polynomial approximation methods for nonlinear activation functions that can substantially reduce complexity without sacrificing performance. The proposed approximation methods were applied to pattern classification problems. Experimental results show that the processing time was reduced by up to 50% without any performance degradations in terms of computer simulation.

AB - Most artificial neural networks use a nonlinear activation function that includes sigmoid and hyperbolic tangent functions. Most artificial networks employ nonlinear functions such as these sigmoid and hyperbolic tangent functions, which incur high complexity costs, particularly during hardware implementation. In this paper, we propose new polynomial approximation methods for nonlinear activation functions that can substantially reduce complexity without sacrificing performance. The proposed approximation methods were applied to pattern classification problems. Experimental results show that the processing time was reduced by up to 50% without any performance degradations in terms of computer simulation.

UR - http://www.scopus.com/inward/record.url?scp=84887934901&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84887934901&partnerID=8YFLogxK

U2 - 10.1117/12.2026666

DO - 10.1117/12.2026666

M3 - Conference contribution

AN - SCOPUS:84887934901

SN - 9780819497215

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Satellite Data Compression, Communications, and Processing IX

PB - SPIE

T2 - Satellite Data Compression, Communications, and Processing IX

Y2 - 26 August 2013 through 27 August 2013

ER -