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

Error correcting memorization learning for noisy training examples.

A Nakashima1, A Hirabayashi, H Ogawa

  • 1Corporate Research and Development Center, Toshiba Corporation, Kawasaki, Japan. akiko.nakashima@toshiba.co.jp

Neural Networks : the Official Journal of the International Neural Network Society
|February 24, 2001
PubMed
Summary

We introduce error correcting memorization learning to prevent overfitting in neural networks. This method effectively suppresses noise in training data without knowing the correct values, enhancing model robustness.

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

Pulmonary abscess caused by Cladosporium cladosporioides after receiving outpatient chemotherapy.

Journal of infection and chemotherapy : official journal of the Japan Society of Chemotherapy·2023
Same author

Five-year point prevalence survey of healthcare-associated infections and antimicrobial use in a Japanese university hospital.

Infection prevention in practice·2021
Same author

Impact of the COVID-19 pandemic on the surveillance of antimicrobial resistance.

The Journal of hospital infection·2021
Same author

Identification of ABCG2 as an Exporter of Uremic Toxin Indoxyl Sulfate in Mice and as a Crucial Factor Influencing CKD Progression.

Scientific reports·2018
Same author

Temperature seasonality during fry out-migration influences the survival of hatchery-reared chum salmon Oncorhynchus keta.

Journal of fish biology·2015
Same author

Characterization of in vitro biotransformation of the new oral anticoagulants, the factor VIIa inhibitors AS1927819-00 and AS1932804-00.

Die Pharmazie·2013

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Neural Networks

Background:

  • Overfitting is a significant challenge in neural network training, leading to poor generalization.
  • Training with noisy data further exacerbates overfitting and reduces model performance.
  • Existing methods often struggle to effectively handle unknown noise in training datasets.

Purpose of the Study:

  • To propose a novel learning method, error correcting memorization learning, to mitigate overfitting caused by noisy training data.
  • To theoretically analyze and clarify the noise suppression mechanism of the proposed method.
  • To investigate the role of redundancy in noise suppression and establish conditions for training inputs.

Main Methods:

  • Error correcting memorization learning is derived from minimizing errors between predicted and true outputs for noisy examples.

Related Experiment Videos

  • The method operates without requiring knowledge of the correct values for the noisy training data.
  • Theoretical analysis is employed to elucidate the noise suppression mechanism and the role of redundancy.

    • Main Results:

    • Error correcting memorization learning effectively suppresses noise in training data.
    • Redundancy within the training input set is identified as crucial for effective noise suppression.
    • Conditions for training inputs to provide the necessary redundancy are established.
    • The relationship between the proposed method and weighted least squares estimation with the Mahalanobis norm is clarified, demonstrating its noise suppression effectiveness.

    Conclusions:

    • Error correcting memorization learning offers a robust solution to overfitting in the presence of noisy training data.
    • The theoretical understanding of noise suppression highlights the importance of data redundancy.
    • The proposed method and its connection to weighted least squares estimation provide valuable insights for developing more resilient machine learning models.