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 MLP training algorithm taking into account known errors on inputs and outputs.

J Svensson1

  • 1Department of Atomic and Molecular Physics, Royal Institute of Technology (KTH), 100 44 Stockholm, Sweden. pjds@ipp.mpg.de

International Journal of Neural Systems
|November 9, 2002
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

Nocturnal glycemic variability and sleep architecture in children and adolescents with type 1 diabetes: A multi-night home sleep study.

Sleep medicine·2026
Same author

Inner southern magnetosphere observation of Mercury via SERENA ion sensors in BepiColombo mission.

Nature communications·2022
Same author

Age-period-cohort modelling of type 1 diabetes incidence rates among children included in the EURODIAB 25-year follow-up study.

Acta diabetologica·2022
Same author

Phase-contrast enhanced synchrotron micro-tomography of human meniscus tissue.

Osteoarthritis and cartilage·2022
Same author

Pre-hospital emergency nurse specialist's experiences in caring for patients with non-specific chief complaints in the ambulance - A qualitative interview study.

International emergency nursing·2022
Same author

Patterns of activity correlate with symptom severity in major depressive disorder patients.

Translational psychiatry·2022
Same journal

Latent Space Projections and Atlases, a Cautionary Tale in Deep Neuroimaging using Autoencoders.

International journal of neural systems·2026
Same journal

Transformer-Based Anomaly Detection for Neurodegenerative Screening in MRI Images.

International journal of neural systems·2026
Same journal

Discrete Wavelet Convolution for Learnable Time-Frequency Representation with Application to Seizure Prediction.

International journal of neural systems·2026
Same journal

Automatic Seizure Detection using Hierarchical Spectral-Temporal Feature Learning with an Imbalance-Aware Transformer.

International journal of neural systems·2026
Same journal

Pyramid Vision Transformer-Enhanced Conformer Network for Epileptic Seizure Recognition Using MultiChannel EEG Signals.

International journal of neural systems·2026
Same journal

A Time-Frequency Decoupled Contrastive Learning Framework for Electroencephalography-Based Parkinson's Disease Diagnosis.

International journal of neural systems·2026
See all related articles

This study introduces a novel training algorithm for Multilayer Perceptron (MLP) networks that incorporates known input and output errors. This approach enhances network regularization and provides a robust method for calculating error bars.

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Neural Networks

Background:

  • Multilayer Perceptron (MLP) networks often face challenges with inherent uncertainties in input and output data.
  • Existing training algorithms may not adequately account for a priori known errors, potentially affecting model accuracy and reliability.

Purpose of the Study:

  • To develop a new training algorithm for MLP networks that explicitly considers known errors in both inputs and outputs.
  • To introduce a novel cost function and derive update formulas for training MLP networks with input/output uncertainties.
  • To establish a method for calculating error bars within a Bayesian framework and compare them to existing distribution width measures.

Main Methods:

  • A new cost function is proposed, based on a linear approximation of the network function over the input distribution.

Related Experiment Videos

  • Update formulas, including the gradient of the new cost function and expressions for the Hessian matrix, are derived for MLP networks.
  • Error bars are calculated using a Bayesian framework, leveraging the derived Hessian matrix.

    • Main Results:

    • The derived error bars are discussed in comparison to the width of the target posterior predictive distribution.
    • The proposed method demonstrates a strong regularizing effect on the MLP network solution when input uncertainties are considered.
    • The algorithm effectively integrates a priori known input and output errors into the MLP training process.

    Conclusions:

    • The novel training algorithm provides a principled way to handle input and output uncertainties in MLP networks.
    • The method offers a robust approach for quantifying uncertainty through derived error bars.
    • Incorporating input uncertainties acts as a significant regularizer, leading to more stable and reliable network solutions.