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

Mutation-based genetic neural network.

Paulito P Palmes1, Taichi Hayasaka, Shiro Usui

  • 1Laboratory for Neuroinformatics, RIKEN Brain Science Institute, Wako City, Saitama 351-0198, Japan. ppalmes@brain.riken.jp

IEEE Transactions on Neural Networks
|June 9, 2005
PubMed
Summary

Mutation-based genetic neural networks (MGNN) offer an alternative to traditional backpropagation for evolving artificial neural networks (ANNs). MGNNs use evolutionary programming for efficient weight learning and dynamic structure adaptation, improving search coverage and generalization.

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

Highly Efficient and Durable Electrocatalysis by a Molecular Catalyst with Long Alkoxyl Chains Immobilized on a Carbon Electrode for Water Oxidation.

ACS applied materials & interfaces·2022
Same author

Proton feedback mediates the cascade of color-opponent signals onto H3 horizontal cells in goldfish retina.

Neuroscience research·2012
Same author

Reprint of: Simulation Platform: a cloud-based online simulation environment.

Neural networks : the official journal of the International Neural Network Society·2011
Same author

PLATO: data-oriented approach to collaborative large-scale brain system modeling.

Neural networks : the official journal of the International Neural Network Society·2011
Same author

Simulation Platform: a cloud-based online simulation environment.

Neural networks : the official journal of the International Neural Network Society·2011
Same author

A model-based theory on the signal transformation for microsaccade generation.

Neural networks : the official journal of the International Neural Network Society·2011

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Computational Neuroscience

Background:

  • Evolving artificial neural networks (ANNs) with evolutionary algorithms (EA) is common for addressing local optima and design challenges.
  • Traditional methods combine backpropagation (BP) for weight learning and EA for architecture search.
  • BP's computational intensity limits EA's search coverage by requiring small population sizes.

Purpose of the Study:

  • To introduce a novel approach, mutation-based genetic neural network (MGNN), to overcome limitations of BP in EA-based ANN evolution.
  • To enable simultaneous dynamic evolution of network structure and weight adaptation.
  • To enhance the search coverage and efficiency of EA for ANN design.

Main Methods:

  • Replaced backpropagation (BP) with a mutation strategy from evolutionary programming (EP) for weight learning in MGNNs.

Related Experiment Videos

  • Utilized an EP-based encoding scheme for flexible fitness function formulation and efficient computation.
  • Implemented a "sliding-window" stopping criterion to monitor and prevent overfitness.

    • Main Results:

    • MGNN demonstrated dynamic evolution of network structure and weight adaptation concurrently.
    • The EP-based encoding facilitated larger population sizes and wider search coverage.
    • Statistical analysis on classification problems confirmed MGNN's good generalization capability.

    Conclusions:

    • MGNN offers an efficient and effective alternative to BP for evolving ANNs, enhancing search capabilities.
    • The flexible fitness function and efficient computation allow for broader exploration of the architecture space.
    • Adaptive scheduling of strategy parameters within MGNN can balance local and global search, further optimizing performance.