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

Improvement of generalization ability for identifying dynamical systems by using universal learning networks.

K Hirasawa1, S Kim, J Hu

  • 1Department of Electrical and Electronic Systems Engineering, Graduate School of Information Science and Electrical Engineering, Kyushu University, Higashiku, Fukuoka, Japan. hirasawa@cig.ees.kyushu-u.ac.jp

Neural Networks : the Official Journal of the International Neural Network Society
|January 5, 2002
PubMed
Summary

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

Evidence of Spin-Interference Effects in Exclusive J/ψ→e^{+}e^{-} Photoproduction in Ultraperipheral Heavy-Ion Collisions.

Physical review letters·2026
Same author

First Observation of Deuteron-Λ Correlations at RHIC.

Physical review letters·2026
Same author

Observation of Charmonium Sequential Suppression in Heavy-Ion Collisions at the Relativistic Heavy Ion Collider.

Physical review letters·2026
Same author

Energy Independence of the Collins Asymmetry in p^{↑}p Collisions.

Physical review letters·2026
Same author

Precision Measurement of Net-Proton-Number Fluctuations in Au+Au Collisions at RHIC.

Physical review letters·2025
Same author

Measurement of Two-Point Energy Correlators within Jets in p+p Collisions at sqrt[s]=200  GeV.

Physical review letters·2025

This study enhances neural network generalization for dynamical systems using second-order derivatives. The Universal Learning Networks (ULNs) method improves model robustness and performance, independent of initial parameters.

Area of Science:

  • Dynamical Systems Modeling
  • Machine Learning
  • Neural Network Theory

Background:

  • Improving the generalization ability of models for dynamical systems is crucial for accurate predictions.
  • Traditional methods often struggle with complex nonlinear dynamics and noise.
  • Universal Learning Networks (ULNs) offer a flexible architecture for modeling dynamical systems.

Purpose of the Study:

  • To enhance the generalization capability of Universal Learning Networks (ULNs) by incorporating second-order derivatives.
  • To develop a robust regularization technique for dynamical system identification.
  • To demonstrate the effectiveness of the proposed method in noisy environments.

Main Methods:

  • Utilizing second-order derivatives of outputs with respect to external inputs for regularization.

Related Experiment Videos

  • Developing a generalized learning algorithm for ULNs that incorporates higher-order derivatives.
  • Implementing a novel regularization term based on second-order derivative information.

    • Main Results:

    • The proposed method significantly improves the generalization ability of neural networks.
    • The resulting networks exhibit robustness even when parts of the trained ULNs are damaged.
    • Performance is independent of initial parameter values, indicating stable learning.

    Conclusions:

    • Second-order derivative-based regularization is effective for improving dynamical system models.
    • The ULN framework with higher-order derivatives provides robust and reliable system identification.
    • This approach offers a significant advancement in modeling complex, noisy dynamical systems.