Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A Sequential Learning Approach for Single Hidden Layer Neural Networks.

A J. Morris1, Jie Zhang

  • 1Centre for Process Analysis, Chemometrics and Control Department of Chemical and Process Engineering, University of Newcastle upon Tyne, Newcastle upon Tyne, U.K.

Neural Networks : the Official Journal of the International Neural Network Society
|March 29, 2003
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

[Distribution and drug resistance spectrum analysis of 2478 clinical bacterial and Candida isolates].

Nan fang yi ke da xue xue bao = Journal of Southern Medical University·2007
Same author

Promoting contraceptive use among unmarried female migrants in one factory in Shanghai: a pilot workplace intervention.

BMC health services research·2007
Same author

Apoptosis related protein 3, an ATRA-upregulated membrane protein arrests the cell cycle at G1/S phase by decreasing the expression of cyclin D1.

Biochemical and biophysical research communications·2007
Same author

Efficient one-pot synthesis of highly substituted pyridin-2(1H)-ones via the Vilsmeier-Haack reaction of 1-acetyl,1-carbamoyl cyclopropanes.

Organic letters·2007
Same author

Effects of Andrographitis Paniculata extracts on the expression of CD40 in endothelial cells.

Journal of Huazhong University of Science and Technology. Medical sciences = Hua zhong ke ji da xue xue bao. Yi xue Ying De wen ban = Huazhong keji daxue xuebao. Yixue Yingdewen ban·2007
Same author

Theoretical and experimental investigations on the size of alginate microspheres prepared by dropping and spraying.

Journal of microencapsulation·2007

This study introduces a sequential orthogonal method for training neural networks. This approach efficiently builds models by adding neurons one by one, ensuring orthogonality for improved accuracy and determining the optimal number of hidden neurons.

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Computational Neuroscience

Background:

  • Single hidden layer neural networks are widely used but determining optimal architecture can be challenging.
  • Existing methods may require extensive trial-and-error for selecting the number of hidden neurons.

Purpose of the Study:

  • To present a novel sequential orthogonal approach for building and training single hidden layer neural networks.
  • To enable automatic determination of the required number of hidden neurons.
  • To enhance network generalization and explore hybrid model development.

Main Methods:

  • A sequential, one-neuron-at-a-time addition strategy is employed.
  • The Gram-Schmidt orthogonalization method is used to ensure new neuron outputs are orthogonal to existing ones.

Related Experiment Videos

  • Hidden layer weights are optimized, and output layer weights are determined via least-squares regression.
  • A regularization factor is incorporated for improved generalization.

    • Main Results:

    • The method successfully determines the necessary number of hidden neurons.
    • Incorporation of a regularization factor enhances network generalization.
    • The approach facilitates the development of hybrid neural networks using mixed neuron types.
    • More accurate models with fewer neurons were developed compared to conventional networks.

    Conclusions:

    • The sequential orthogonal training method provides an effective way to build and train neural networks.
    • This approach leads to more accurate and parsimonious models, especially when using hybrid architectures.
    • The technique was validated on three non-linear modeling problems.