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

An efficient parameterization of dynamic neural networks for nonlinear system identification.

Victor M Becerra, Freddy R Garces, Slawomir J Nasuto

    IEEE Transactions on Neural Networks
    |August 27, 2005
    PubMed
    Summary

    This study introduces an efficient parameterization for dynamic neural networks (DNNs), simplifying nonlinear system identification. This approach reduces the number of parameters, leading to more parsimonious models and easier training.

    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

    North-South asymmetry of Parkinson's disease mortality in Brazil between 2009 and 2023: a spatial analysis.

    Scientific reports·2026
    Same author

    Takotsubo Syndrome in the Setting of Pacemaker Implantation: Results From a Multicenter National Prospective Registry.

    Reviews in cardiovascular medicine·2025
    Same author

    Machine Learning-Enhanced Attribute-Based Authentication for Secure IoT Access Control.

    Sensors (Basel, Switzerland)·2025
    Same author

    Brain tumor classification using MRI images and deep learning techniques.

    PloS one·2025
    Same author

    Alpha-Oscillatory Current Application Impacts Prospective Remembering Through Strategic Monitoring.

    Psychophysiology·2025
    Same author

    Quantitative metrics for evaluating surgical dexterity using virtual reality simulations.

    PloS one·2025

    Area of Science:

    • Engineering
    • Computer Science
    • Control Theory

    Background:

    • Dynamic neural networks (DNNs), also known as recurrent neural networks, are widely applied in nonlinear system identification.
    • Training DNNs can be complex due to the large number of parameters requiring adjustment.
    • Parsimonious models are desirable for efficient and robust system identification.

    Purpose of the Study:

    • To introduce an efficient parameterization for a class of dynamic neural networks.
    • To simplify the training process for DNNs used in nonlinear system identification.
    • To develop more parsimonious dynamic neural network models.

    Main Methods:

    • Developed a novel parameterization for dynamic neural networks based on approximation theory.
    • Utilized approximation theory concerning the ability of DNNs to approximate finite trajectories of nonautonomous systems.

    Related Experiment Videos

  • Applied the proposed parameterization to a nonlinear magnetic levitation system model.

    • Main Results:

    • The proposed parameterization significantly reduces the number of adjustable parameters in dynamic neural networks.
    • This reduction simplifies the training problem, leading to more parsimonious models.
    • Demonstrated the effectiveness through a numerical example using magnetic levitation system data.

    Conclusions:

    • The novel parameterization offers an efficient approach to dynamic neural network design for nonlinear system identification.
    • Reduced parameterization leads to improved model parsimony and training efficiency.
    • The method is validated by successful application to a complex nonlinear system.