Kernel and range approach to analytic network learning

Research output: Contribution to journalArticle

Abstract

A novel learning approach for a composite function that can be written in the form of a matrix system of linear equations is introduced in this paper. This learning approach, which is gradient-free, is grounded upon the observation that solving the system of linear equations by manipulating the kernel and the range projection spaces using the Moore–Penrose inversion boils down to an approximation in the least squares error sense. In view of the heavy dependence on computation of the pseudoinverse, a simplification method is proposed. The learning approach is applied to learn a multilayer feedforward neural network with full weight connections. The numerical experiments on learning both synthetic and benchmark data sets not only validate the feasibility but also depict the performance of the proposed formulation.

Original languageEnglish
Pages (from-to)20-28
Number of pages9
JournalInternational Journal of Networked and Distributed Computing
Volume7
Issue number1
DOIs
Publication statusPublished - 2018 Dec 1

Fingerprint

Linear equations
Feedforward neural networks
Multilayer neural networks
Composite materials
Experiments

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Computer Networks and Communications

Cite this

@article{ef461ddb661044e1911954a4774f899b,
title = "Kernel and range approach to analytic network learning",
abstract = "A novel learning approach for a composite function that can be written in the form of a matrix system of linear equations is introduced in this paper. This learning approach, which is gradient-free, is grounded upon the observation that solving the system of linear equations by manipulating the kernel and the range projection spaces using the Moore–Penrose inversion boils down to an approximation in the least squares error sense. In view of the heavy dependence on computation of the pseudoinverse, a simplification method is proposed. The learning approach is applied to learn a multilayer feedforward neural network with full weight connections. The numerical experiments on learning both synthetic and benchmark data sets not only validate the feasibility but also depict the performance of the proposed formulation.",
author = "Toh, {Kar Ann}",
year = "2018",
month = "12",
day = "1",
doi = "10.2991/ijndc.2018.7.1.3",
language = "English",
volume = "7",
pages = "20--28",
journal = "International Journal of Networked and Distributed Computing",
issn = "2211-7938",
publisher = "Atlantis Press",
number = "1",

}

Kernel and range approach to analytic network learning. / Toh, Kar Ann.

In: International Journal of Networked and Distributed Computing, Vol. 7, No. 1, 01.12.2018, p. 20-28.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Kernel and range approach to analytic network learning

AU - Toh, Kar Ann

PY - 2018/12/1

Y1 - 2018/12/1

N2 - A novel learning approach for a composite function that can be written in the form of a matrix system of linear equations is introduced in this paper. This learning approach, which is gradient-free, is grounded upon the observation that solving the system of linear equations by manipulating the kernel and the range projection spaces using the Moore–Penrose inversion boils down to an approximation in the least squares error sense. In view of the heavy dependence on computation of the pseudoinverse, a simplification method is proposed. The learning approach is applied to learn a multilayer feedforward neural network with full weight connections. The numerical experiments on learning both synthetic and benchmark data sets not only validate the feasibility but also depict the performance of the proposed formulation.

AB - A novel learning approach for a composite function that can be written in the form of a matrix system of linear equations is introduced in this paper. This learning approach, which is gradient-free, is grounded upon the observation that solving the system of linear equations by manipulating the kernel and the range projection spaces using the Moore–Penrose inversion boils down to an approximation in the least squares error sense. In view of the heavy dependence on computation of the pseudoinverse, a simplification method is proposed. The learning approach is applied to learn a multilayer feedforward neural network with full weight connections. The numerical experiments on learning both synthetic and benchmark data sets not only validate the feasibility but also depict the performance of the proposed formulation.

UR - http://www.scopus.com/inward/record.url?scp=85061499838&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85061499838&partnerID=8YFLogxK

U2 - 10.2991/ijndc.2018.7.1.3

DO - 10.2991/ijndc.2018.7.1.3

M3 - Article

AN - SCOPUS:85061499838

VL - 7

SP - 20

EP - 28

JO - International Journal of Networked and Distributed Computing

JF - International Journal of Networked and Distributed Computing

SN - 2211-7938

IS - 1

ER -