MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers

Hao Wang; Andrew Chi Sing Leung; John Sum

doi:10.1007/978-3-030-04179-3_59

MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers

Department of Electrical Engineering

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 12 - Chapter in an edited book (Author) › peer-review

Abstract

In the implementation of a neural network, some imperfect issues, such as precision error and thermal noise, always exist. They can be modeled as multiplicative noise. This paper studies the problem of training RBF network and selecting centers under multiplicative noise. We devise a noise resistant training algorithm based on the alternating direction method of multipliers (ADMM) framework and the minimax concave penalty (MCP) function. Our algorithm first uses all training samples to create the RBF nodes. Afterwards, we derive the training objective function that can tolerate to the existence of noise. Finally, we add a MCP term to the objective function. We then apply the ADMM framework to minimize the modified objective function. During training, the MCP term has an ability to make some unimportant RBF weights to zero. Hence training and RBF node selection can be done at the same time. The proposed algorithm is called the ADMM-MCP algorithm. Also, we present the convergent properties of the ADMM-MCP algorithm. From the simulation result, the ADMM-MCP algorithm is better than many other RBF training algorithms under weight/node noise situation.

Original language	English
Title of host publication	Neural Information Processing
Subtitle of host publication	25th International Conference, ICONIP 2018, Proceedings
Editors	Long Cheng, Andrew Chi Sing Leung, Seiichi Ozawa
Publisher	Springer, Cham
Pages	668-679
Volume	2
ISBN (Electronic)	9783030041793
ISBN (Print)	9783030041786
DOIs	https://doi.org/10.1007/978-3-030-04179-3_59
Publication status	Published - Dec 2018
Event	25th International Conference on Neural Information Processing (ICONIP 2018) - Sokha Siem Reap Resort & Convention Center, Siem Reap, Cambodia Duration: 13 Dec 2018 → 16 Dec 2018 https://conference.cs.cityu.edu.hk/iconip/

Publication series

Name	Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues)
Publisher	Springer, Cham
Volume	11302
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	25th International Conference on Neural Information Processing (ICONIP 2018)
Abbreviated title	ICONIP 2018
Place	Cambodia
City	Siem Reap
Period	13/12/18 → 16/12/18
Internet address	https://conference.cs.cityu.edu.hk/iconip/

Research Keywords

ADMM
Center selection
MCP
Multiplicative noise
RBF

Access Output

10.1007/978-3-030-04179-3_59

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

Wang, H., Leung, A. C. S., & Sum, J. (2018). MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers. In L. Cheng, A. C. S. Leung, & S. Ozawa (Eds.), Neural Information Processing: 25th International Conference, ICONIP 2018, Proceedings (Vol. 2, pp. 668-679). (Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues); Vol. 11302). Springer, Cham. https://doi.org/10.1007/978-3-030-04179-3_59

Wang, Hao ; Leung, Andrew Chi Sing ; Sum, John. / MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers. Neural Information Processing: 25th International Conference, ICONIP 2018, Proceedings. editor / Long Cheng ; Andrew Chi Sing Leung ; Seiichi Ozawa. Vol. 2 Springer, Cham, 2018. pp. 668-679 (Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues)).

@inbook{fa9dd9b7f94c422fbf4b32b299ee71f2,

title = "MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers",

abstract = "In the implementation of a neural network, some imperfect issues, such as precision error and thermal noise, always exist. They can be modeled as multiplicative noise. This paper studies the problem of training RBF network and selecting centers under multiplicative noise. We devise a noise resistant training algorithm based on the alternating direction method of multipliers (ADMM) framework and the minimax concave penalty (MCP) function. Our algorithm first uses all training samples to create the RBF nodes. Afterwards, we derive the training objective function that can tolerate to the existence of noise. Finally, we add a MCP term to the objective function. We then apply the ADMM framework to minimize the modified objective function. During training, the MCP term has an ability to make some unimportant RBF weights to zero. Hence training and RBF node selection can be done at the same time. The proposed algorithm is called the ADMM-MCP algorithm. Also, we present the convergent properties of the ADMM-MCP algorithm. From the simulation result, the ADMM-MCP algorithm is better than many other RBF training algorithms under weight/node noise situation.",

keywords = "ADMM, Center selection, MCP, Multiplicative noise, RBF",

author = "Hao Wang and Leung, \{Andrew Chi Sing\} and John Sum",

year = "2018",

month = dec,

doi = "10.1007/978-3-030-04179-3\_59",

language = "English",

isbn = "9783030041786",

volume = "2",

series = "Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues)",

publisher = "Springer, Cham",

pages = "668--679",

editor = "Long Cheng and Leung, \{Andrew Chi Sing\} and Seiichi Ozawa",

booktitle = "Neural Information Processing",

note = "25th International Conference on Neural Information Processing (ICONIP 2018), ICONIP 2018 ; Conference date: 13-12-2018 Through 16-12-2018",

url = "https://conference.cs.cityu.edu.hk/iconip/",

}

Wang, H , Leung, ACS & Sum, J 2018, MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers. in L Cheng, ACS Leung & S Ozawa (eds), Neural Information Processing: 25th International Conference, ICONIP 2018, Proceedings. vol. 2, Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues), vol. 11302, Springer, Cham, pp. 668-679, 25th International Conference on Neural Information Processing (ICONIP 2018), Siem Reap, Cambodia, 13/12/18. https://doi.org/10.1007/978-3-030-04179-3_59

MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers. / Wang, Hao ; Leung, Andrew Chi Sing; Sum, John.
Neural Information Processing: 25th International Conference, ICONIP 2018, Proceedings. ed. / Long Cheng; Andrew Chi Sing Leung; Seiichi Ozawa. Vol. 2 Springer, Cham, 2018. p. 668-679 (Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues); Vol. 11302).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 12 - Chapter in an edited book (Author) › peer-review

TY - CHAP

T1 - MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers

AU - Wang, Hao

AU - Leung, Andrew Chi Sing

AU - Sum, John

PY - 2018/12

Y1 - 2018/12

N2 - In the implementation of a neural network, some imperfect issues, such as precision error and thermal noise, always exist. They can be modeled as multiplicative noise. This paper studies the problem of training RBF network and selecting centers under multiplicative noise. We devise a noise resistant training algorithm based on the alternating direction method of multipliers (ADMM) framework and the minimax concave penalty (MCP) function. Our algorithm first uses all training samples to create the RBF nodes. Afterwards, we derive the training objective function that can tolerate to the existence of noise. Finally, we add a MCP term to the objective function. We then apply the ADMM framework to minimize the modified objective function. During training, the MCP term has an ability to make some unimportant RBF weights to zero. Hence training and RBF node selection can be done at the same time. The proposed algorithm is called the ADMM-MCP algorithm. Also, we present the convergent properties of the ADMM-MCP algorithm. From the simulation result, the ADMM-MCP algorithm is better than many other RBF training algorithms under weight/node noise situation.

AB - In the implementation of a neural network, some imperfect issues, such as precision error and thermal noise, always exist. They can be modeled as multiplicative noise. This paper studies the problem of training RBF network and selecting centers under multiplicative noise. We devise a noise resistant training algorithm based on the alternating direction method of multipliers (ADMM) framework and the minimax concave penalty (MCP) function. Our algorithm first uses all training samples to create the RBF nodes. Afterwards, we derive the training objective function that can tolerate to the existence of noise. Finally, we add a MCP term to the objective function. We then apply the ADMM framework to minimize the modified objective function. During training, the MCP term has an ability to make some unimportant RBF weights to zero. Hence training and RBF node selection can be done at the same time. The proposed algorithm is called the ADMM-MCP algorithm. Also, we present the convergent properties of the ADMM-MCP algorithm. From the simulation result, the ADMM-MCP algorithm is better than many other RBF training algorithms under weight/node noise situation.

KW - ADMM

KW - Center selection

KW - MCP

KW - Multiplicative noise

KW - RBF

UR - https://www.scopus.com/pages/publications/85059081010

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85059081010&origin=recordpage

U2 - 10.1007/978-3-030-04179-3_59

DO - 10.1007/978-3-030-04179-3_59

M3 - RGC 12 - Chapter in an edited book (Author)

SN - 9783030041786

VL - 2

T3 - Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues)

SP - 668

EP - 679

BT - Neural Information Processing

A2 - Cheng, Long

A2 - Leung, Andrew Chi Sing

A2 - Ozawa, Seiichi

PB - Springer, Cham

T2 - 25th International Conference on Neural Information Processing (ICONIP 2018)

Y2 - 13 December 2018 through 16 December 2018

ER -

Wang H , Leung ACS, Sum J. MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers. In Cheng L, Leung ACS, Ozawa S, editors, Neural Information Processing: 25th International Conference, ICONIP 2018, Proceedings. Vol. 2. Springer, Cham. 2018. p. 668-679. (Lecture Notes in Computer Science (including subseries Theoretical Computer Science and General Issues)). Epub 2018 Nov 18. doi: 10.1007/978-3-030-04179-3_59

MCP Based Noise Resistant Algorithm for Training RBF Networks and Selecting Centers

Abstract

Publication series

Conference

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this