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

Effective learning in recurrent max-min neural networks.

Kia Fock Loe1, Loo Nin Teow

  • 1Department of Information Systems and Computer Science, National University of Singapore, Singapore, Singapore

Neural Networks : the Official Journal of the International Neural Network Society
|March 29, 2003
PubMed
Summary

This study introduces a new gradient descent method for max-min neural networks, enabling effective learning and improved grammatical inference. The novel approach enhances learning speed and generalization, allowing for straightforward extraction of deterministic finite automata.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same journal

A Multi-Task Learning Framework with Physics Embedded for Signal Reconstruction and State Prediction in Underground Multi-Robot Systems.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Multi-scale Fourier network for all-in-one adverse weather removal of power line aerial image.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

C<sup>2</sup>LNet: Fusing infrared and visible image via a cross-modality complementarity learning network.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Upper bounds for local learning coefficients of three-layer neural networks.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Any-step dynamics model improves future predictions for offline reinforcement learning via hierarchical roll-out.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Dynamical behavior and applications of fractional-order bicyclic crossed memristive neural networks with Neimark-Sacker bifurcation: Synchronization and image encryption.

Neural networks : the official journal of the International Neural Network Society·2026

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Neural Networks

Background:

  • Max and min operations bridge symbolic and real-valued domains.
  • Neural networks utilizing max-min activation functions are gaining interest.
  • Non-strict differentiability of max-min functions poses learning challenges.

Purpose of the Study:

  • To develop a mathematically sound learning method for max-min neural networks.
  • To propose a novel recurrent max-min neural network model.
  • To demonstrate the model's effectiveness in grammatical inference.

Main Methods:

  • Utilized Fourier convergence analysis of side-derivatives to derive a gradient descent technique.
  • Developed a novel recurrent max-min neural network architecture.

Related Experiment Videos

  • Trained the model for grammatical inference and compared it with recurrent sigmoidal networks.

    • Main Results:

    • The proposed gradient descent technique enables effective learning in max-min neural networks.
    • The novel recurrent max-min network shows superior learning speed and generalization compared to sigmoidal networks.
    • The trained model facilitates straightforward extraction of a deterministic finite automaton (DFA).

    Conclusions:

    • The developed gradient descent method overcomes differentiability issues in max-min networks.
    • Max-min neural networks offer advantages in learning speed, generalization, and model interpretability (DFA extraction).
    • This research validates the efficacy of max-min neural networks for tasks like grammatical inference.