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

Feedback stabilization using two-hidden-layer nets.

E D Sontag1

  • 1Dept. of Math., Rutgers Univ., New Brunswick, NJ.

IEEE Transactions on Neural Networks
|January 1, 1992
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

Monotone bifurcation graphs.

Journal of biological dynamics·2012
Same author

Remarks on feedforward circuits, adaptation, and pulse memory.

IET systems biology·2009
Same author

Proximity of intracellular regulatory networks to monotone systems.

IET systems biology·2008
Same author

Some new directions in control theory inspired by systems biology.

Systems biology·2006
Same author

Methods of robustness analysis for Boolean models of gene control networks.

Systems biology·2006
Same author

Parameter estimation in models combining signal transduction and metabolic pathways: the dependent input approach.

Systems biology·2006
Same journal

Universal perceptron and DNA-like learning algorithm for binary neural networks: LSBF and PBF implementations.

IEEE transactions on neural networks·2013
Same journal

Guest editorial: special section on white box nonlinear prediction models.

IEEE transactions on neural networks·2011
Same journal

Data-based fault-tolerant control of high-speed trains with traction/braking notch nonlinearities and actuator failures.

IEEE transactions on neural networks·2011
Same journal

Guest editorial: special section on data-based control, modeling, and optimization.

IEEE transactions on neural networks·2011
Same journal

Neural network-based multiple robot simultaneous localization and mapping.

IEEE transactions on neural networks·2011
Same journal

Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems.

IEEE transactions on neural networks·2011
See all related articles

Two hidden layers are necessary for specific problems, particularly inverse problems like control systems and inverse kinematics, which differ from direct function approximation tasks. This contrasts with general approximation theorems suggesting one layer might suffice.

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Computational Neuroscience

Background:

  • Feedforward neural networks with linear threshold units are fundamental in machine learning.
  • Approximation theorems suggest universal representational power, potentially implying limited need for deep architectures.
  • Understanding the precise layer requirements for different problem types is crucial for effective model design.

Purpose of the Study:

  • To compare the representational capabilities of one-hidden-layer versus two-hidden-layer feedforward neural networks.
  • To identify specific problem classes where two hidden layers offer advantages over one.
  • To elucidate the theoretical underpinnings of these differences, particularly concerning direct and inverse problems.

Main Methods:

Related Experiment Videos

  • Comparative analysis of network architectures (one vs. two hidden layers).
  • Theoretical examination of function approximation capabilities for direct and inverse problems.
  • Demonstration of stabilization capabilities for nonlinear control systems.

    • Main Results:

    • Certain problems necessitate two hidden layers, contrary to expectations from approximation theorems.
    • The distinction lies in classifying problems as direct (function approximation) or inverse (one-sided inverse approximation).
    • Two hidden layers can stabilize nonlinear control systems, whereas one hidden layer generally cannot.

    Conclusions:

    • The number of hidden layers required is critically dependent on the problem type, specifically distinguishing between direct and inverse problems.
    • Two-hidden-layer networks offer enhanced capabilities for inverse problems, such as in control theory and inverse kinematics.
    • This finding has implications for designing neural network architectures for complex tasks beyond simple function approximation.