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

A numerical study on learning curves in stochastic multilayer feedforward networks

K R Müller1, M Finke, N Murata

  • 1Department of Mathematical Engineering and Inf. Physics, University of Tokyo, Japan.

Neural Computation
|July 1, 1996
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

Outbreak of <i>Serratia marcescens</i> mastitis linked to teat spray contamination in a New Zealand dairy herd.

New Zealand veterinary journal·2026
Same author

Can we estimate herd-level prevalence of lameness in dairy cow herds kept at pasture by sampling part of the herd?

New Zealand veterinary journal·2025
Same author

Repeatability of whole herd lameness scoring: an analysis of a New Zealand dataset.

New Zealand veterinary journal·2024
Same author

Farm-level risk factors and treatment protocols for lameness in New Zealand dairy cattle.

New Zealand veterinary journal·2024
Same author

Prevalence of lameness on pasture-based New Zealand dairy farms: An observational study.

Preventive veterinary medicine·2023
Same author

Impact of control measures including decolonization and hand hygiene for orthopaedic surgical site infection caused by MRSA at a Japanese tertiary-care hospital.

The Journal of hospital infection·2023

Large-scale simulations of neural networks confirm Amari

Area of Science:

  • Computational neuroscience
  • Machine learning theory

Background:

  • Amari et al. proposed universal asymptotic scaling laws for neural networks.
  • Understanding generalization error is crucial for network performance.

Purpose of the Study:

  • Investigate Amari's scaling laws in large-scale simulations.
  • Analyze generalization error in stochastic multilayer feedforward networks.
  • Examine the impact of training data size on scaling laws.

Main Methods:

  • Conducted large-scale simulations using a CM5 supercomputer.
  • Trained small stochastic multilayer feedforward networks with backpropagation.
  • Analyzed generalization error as a function of training patterns (t).

Main Results:

Related Experiment Videos

  • Observed 1/t scaling of generalization error for large t, matching predictions.
  • Found a faster 1/t^2 scaling for medium t, explained by higher-order corrections.
  • Demonstrated a drastic change in scaling laws during the transition from overfitting to effective learning for small t.

Conclusions:

  • Validated Amari's asymptotic scaling laws for generalization error.
  • Highlighted the importance of higher-order corrections for medium training data regimes.
  • Characterized the critical transition in network learning dynamics as data size varies.