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

Confidence in prediction by neural networks.

L Ein-Dor1, I Kanter

  • 1Minerva Center and Department of Physics, Bar-Ilan University, Ramat-Gan 52900, Israel.

Physical Review. E, Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics
|April 24, 2002
PubMed
Summary

This study examines how trained neural networks can provide confidence scores for predictions, enhancing reliability assessment. We analyzed perceptrons and estimated information gain using entropy measurements.

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

Loss of lamin A function increases chromatin dynamics in the nuclear interior.

Nature communications·2015
Same author

Coupled lasers: phase versus chaos synchronization.

Optics letters·2013
Same author

Random symmetry breaking and freezing in chaotic networks.

Physical review. E, Statistical, nonlinear, and soft matter physics·2012
Same author

Sensitivity of global network dynamics to local parameters versus motif structure in a cortexlike neuronal model.

Physical review. E, Statistical, nonlinear, and soft matter physics·2012
Same author

Synchronization in small networks of time-delay coupled chaotic diode lasers.

Optics express·2012
Same author

Coexistence of exponentially many chaotic spin-glass attractors.

Physical review. E, Statistical, nonlinear, and soft matter physics·2012

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Information Theory

Background:

  • Assessing the reliability of predictions from trained neural networks is crucial for trustworthy AI.
  • Current methods often lack a quantitative measure of prediction confidence.
  • Understanding the information gain from confidence scores can improve model evaluation.

Purpose of the Study:

  • To investigate the capability of trained networks to generate confidence scores for their predictions.
  • To analytically examine perceptrons with discrete and continuous outputs in this context.
  • To quantify the information gain provided by confidence scores using entropy.

Main Methods:

  • Analytical examination of a perceptron model.
  • Derivation of results for both Gibbs and Bayes scenarios.
  • Estimation of information gain via various entropy measurements.

Main Results:

  • Demonstrated that perceptrons can assign confidence numbers to predictions.
  • Quantified the reliability information provided by these confidence scores.
  • Showcased the utility of entropy measurements in estimating information gain.

Conclusions:

  • Confidence scores from trained networks offer a reliable measure of prediction certainty.
  • The analytical framework provides a method for evaluating confidence in neural network outputs.
  • Information gain estimation using entropy is effective for assessing the value of confidence scores.

Related Experiment Videos