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

Jacobian conditioning analysis for model validation.

Isabelle Rivals1, Léon Personnaz

  • 1Equipe de Statistique Appliquée, Ecole Supérieure de Physique et de Chimie Industrielles, Paris, France. Isabelle.Rivals@espci.fr

Neural Computation
|March 10, 2004
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

Association between dyspnea and cognitive task performance in amyotrophic lateral sclerosis: An exploratory study.

Respiratory physiology & neurobiology·2026
Same author

Differential biological responses to dyspnea-inducing experimental respiratory challenges in healthy humans.

Journal of applied physiology (Bethesda, Md. : 1985)·2026
Same author

SSVEP-based brain-computer interface enabling graded dyspnoea self-report: proof-of-concept study in healthy volunteers.

Journal of neuroengineering and rehabilitation·2026
Same author

Physiological effects of nasal high flow therapy during exercise in patients with chronic obstructive pulmonary disease: A crossover randomised controlled trial.

Pulmonology·2025
Same author

Physiological effects of nasal high flow therapy during exercise in patients with chronic obstructive pulmonary disease: A crossover randomised controlled trial.

Pulmonology·2025
Same author

Medical hypnosis mitigates laboratory dyspnoea in healthy humans: a randomised, controlled experimental trial.

The European respiratory journal·2024

Jacobian matrix conditioning is crucial for validating neural network models. This study questions using leverage values, instead of condition numbers, to identify models prone to overfitting or inaccurate confidence intervals.

Area of Science:

  • Machine Learning
  • Numerical Analysis
  • Statistical Modeling

Background:

  • Neural network model validation is critical for reliable predictions and interpretations.
  • Previous work suggested discarding models likely to overfit or yield inaccurate confidence intervals.
  • Existing methods propose using Jacobian matrix properties for model selection.

Purpose of the Study:

  • To emphasize the importance of Jacobian matrix conditioning in neural network model validation.
  • To critically evaluate the use of leverage values versus Jacobian condition numbers for model selection.
  • To address discrepancies in model validation approaches proposed by Monari and Dreyfus (2002).

Main Methods:

  • Analysis of Jacobian matrix conditioning and its relationship to model overfitting.

Related Experiment Videos

  • Investigation of leverage values (diagonal elements of the hat matrix) as a model validation metric.
  • Theoretical examination of the hat matrix definition and its properties.
  • Comparison of Jacobian condition numbers and leverage values for identifying ill-conditioned networks.

    • Main Results:

    • Jacobian matrix conditioning is a more reliable indicator for model validation than leverage values.
    • Leverage values may appear within theoretical bounds even for ill-conditioned networks, masking potential issues.
    • Accurate estimation of confidence intervals can be compromised in networks with ill-conditioned Jacobians, irrespective of leverage values.
    • The hat matrix is theoretically defined regardless of the Jacobian matrix's rank.

    Conclusions:

    • Jacobian matrix condition number is a superior metric for discarding neural network models prone to overfitting or estimation inaccuracies.
    • Relying solely on leverage values for model validation can be misleading, as demonstrated by their behavior in ill-conditioned networks.
    • Accurate confidence interval estimation necessitates careful consideration of Jacobian matrix conditioning.