Predictive performance of "highly complex" artificial neural networks

A. Tompos, József L. Margitfalvi, E. Tfirst, K. Heberger

Research output: Contribution to journalArticle

21 Citations (Scopus)

Abstract

The effectiveness and the "indeterminacy" of artificial neural networks (ANNs) are discussed. In the way of determination of parameters of different models a clear distinction should be made between two approaches: (i) parameter estimation (classical approach) and (ii) ANN approach with the aim of prediction (mainly interpolation). In the classical approach different regression procedures are used, while in latter one the so-called "training" using back propagation algorithm is applied. The latter approach is widely employed in combinatorial materials science for "information mining" purposes. Parameters obtained in classical approach should have a definite physical meaning. It also means that in the absence of a presumed hypothetical model the classical regression procedures cannot be applied. The advantage of the usage of ANNs is that it does not require a casual model for the system investigated. However, parameters obtained by ANN have no physical meaning. Generally, in ANNs the number of parameters to be determined is significantly higher than the number of data in the training set. It has to be emphasized that in ANN approach the good predictive ability has superior importance to physical meaning of parameters.

Original languageEnglish
Pages (from-to)90-93
Number of pages4
JournalApplied Catalysis A: General
Volume324
Issue number1-2
DOIs
Publication statusPublished - May 17 2007

Fingerprint

Neural networks
Backpropagation algorithms
Materials science
Parameter estimation
Interpolation

Keywords

  • Artificial neural networks
  • Catalyst library design
  • Chemometrics
  • Combinatorial catalysis
  • Genetic algorithm
  • Holographic research strategy
  • Overfit
  • Visualization

ASJC Scopus subject areas

  • Catalysis
  • Process Chemistry and Technology

Cite this

Predictive performance of "highly complex" artificial neural networks. / Tompos, A.; Margitfalvi, József L.; Tfirst, E.; Heberger, K.

In: Applied Catalysis A: General, Vol. 324, No. 1-2, 17.05.2007, p. 90-93.

Research output: Contribution to journalArticle

@article{04ce8b5e52024b73b9ccc2247876ddd4,
title = "Predictive performance of {"}highly complex{"} artificial neural networks",
abstract = "The effectiveness and the {"}indeterminacy{"} of artificial neural networks (ANNs) are discussed. In the way of determination of parameters of different models a clear distinction should be made between two approaches: (i) parameter estimation (classical approach) and (ii) ANN approach with the aim of prediction (mainly interpolation). In the classical approach different regression procedures are used, while in latter one the so-called {"}training{"} using back propagation algorithm is applied. The latter approach is widely employed in combinatorial materials science for {"}information mining{"} purposes. Parameters obtained in classical approach should have a definite physical meaning. It also means that in the absence of a presumed hypothetical model the classical regression procedures cannot be applied. The advantage of the usage of ANNs is that it does not require a casual model for the system investigated. However, parameters obtained by ANN have no physical meaning. Generally, in ANNs the number of parameters to be determined is significantly higher than the number of data in the training set. It has to be emphasized that in ANN approach the good predictive ability has superior importance to physical meaning of parameters.",
keywords = "Artificial neural networks, Catalyst library design, Chemometrics, Combinatorial catalysis, Genetic algorithm, Holographic research strategy, Overfit, Visualization",
author = "A. Tompos and Margitfalvi, {J{\'o}zsef L.} and E. Tfirst and K. Heberger",
year = "2007",
month = "5",
day = "17",
doi = "10.1016/j.apcata.2007.02.052",
language = "English",
volume = "324",
pages = "90--93",
journal = "Applied Catalysis A: General",
issn = "0926-860X",
publisher = "Elsevier",
number = "1-2",

}

TY - JOUR

T1 - Predictive performance of "highly complex" artificial neural networks

AU - Tompos, A.

AU - Margitfalvi, József L.

AU - Tfirst, E.

AU - Heberger, K.

PY - 2007/5/17

Y1 - 2007/5/17

N2 - The effectiveness and the "indeterminacy" of artificial neural networks (ANNs) are discussed. In the way of determination of parameters of different models a clear distinction should be made between two approaches: (i) parameter estimation (classical approach) and (ii) ANN approach with the aim of prediction (mainly interpolation). In the classical approach different regression procedures are used, while in latter one the so-called "training" using back propagation algorithm is applied. The latter approach is widely employed in combinatorial materials science for "information mining" purposes. Parameters obtained in classical approach should have a definite physical meaning. It also means that in the absence of a presumed hypothetical model the classical regression procedures cannot be applied. The advantage of the usage of ANNs is that it does not require a casual model for the system investigated. However, parameters obtained by ANN have no physical meaning. Generally, in ANNs the number of parameters to be determined is significantly higher than the number of data in the training set. It has to be emphasized that in ANN approach the good predictive ability has superior importance to physical meaning of parameters.

AB - The effectiveness and the "indeterminacy" of artificial neural networks (ANNs) are discussed. In the way of determination of parameters of different models a clear distinction should be made between two approaches: (i) parameter estimation (classical approach) and (ii) ANN approach with the aim of prediction (mainly interpolation). In the classical approach different regression procedures are used, while in latter one the so-called "training" using back propagation algorithm is applied. The latter approach is widely employed in combinatorial materials science for "information mining" purposes. Parameters obtained in classical approach should have a definite physical meaning. It also means that in the absence of a presumed hypothetical model the classical regression procedures cannot be applied. The advantage of the usage of ANNs is that it does not require a casual model for the system investigated. However, parameters obtained by ANN have no physical meaning. Generally, in ANNs the number of parameters to be determined is significantly higher than the number of data in the training set. It has to be emphasized that in ANN approach the good predictive ability has superior importance to physical meaning of parameters.

KW - Artificial neural networks

KW - Catalyst library design

KW - Chemometrics

KW - Combinatorial catalysis

KW - Genetic algorithm

KW - Holographic research strategy

KW - Overfit

KW - Visualization

UR - http://www.scopus.com/inward/record.url?scp=34247390731&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=34247390731&partnerID=8YFLogxK

U2 - 10.1016/j.apcata.2007.02.052

DO - 10.1016/j.apcata.2007.02.052

M3 - Article

VL - 324

SP - 90

EP - 93

JO - Applied Catalysis A: General

JF - Applied Catalysis A: General

SN - 0926-860X

IS - 1-2

ER -