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

Combining expert neural networks using reinforcement feedback for learning primitive grasping behavior.

Medhat A Moussa1

  • 1School of Engineering, University of Guelph, Guelph, ON, Canada NIG 2W1. mmoussa@uoguelph.ca

IEEE Transactions on Neural Networks
|September 24, 2004
PubMed
Summary

This study introduces a novel mixture of experts architecture that dynamically expands and combines neural networks using reinforcement learning. This approach effectively handles complex many-to-many mappings, demonstrated in robot grasping tasks.

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

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

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Robotics

Background:

  • Complex real-world problems often require approximating many-to-many mappings.
  • Existing methods for combining expert models may lack flexibility in adapting to unknown structures.

Purpose of the Study:

  • To present a novel architecture for combining a mixture of experts.
  • To enable dynamic expansion of experts during training without prior structural knowledge.
  • To utilize reinforcement feedback for guiding the combining and expansion processes.

Main Methods:

  • Developed a mixture of experts architecture with dynamic expansion capabilities.
  • Employed reinforcement learning to guide the combining and expansion of experts.
  • Tested the architecture on a robot grasping task involving a database of 28 objects in a simulated environment.

Related Experiment Videos

Main Results:

  • The algorithm successfully learned to approximate the many-to-many mapping required for robot grasping.
  • Demonstrated dynamic combination and expansion of a mixture of neural networks during training.
  • Achieved effective learning in a simulated environment, outperforming non-learning approaches in comparisons.

Conclusions:

  • The proposed architecture offers a flexible and adaptive solution for problems requiring many-to-many mapping approximation.
  • Dynamic expansion and reinforcement learning provide an effective mechanism for complex learning tasks.
  • The approach shows promise for robotic applications, such as object grasping.