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

DMP3: a dynamic multilayer perceptron construction algorithm.

T L Andersen1, T R Martinez

  • 1Computer Science Department, Brigham Young University, Provo, Utah 84604, USA. tim@axon.cs.byu.edu

International Journal of Neural Systems
|November 25, 2003
PubMed
Summary

Dynamic Multilayer Perceptron 3 (DMP3) is a novel constructive training method for multilayer perceptrons (MLPs). DMP3 outperforms existing machine learning algorithms in generalization performance across various datasets.

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

Type 2 diabetes patients exhibit delayed but coupled bone remodelling, maintaining cortical porosity comparable to healthy controls: A histomorphometric analysis of trans-iliac bone biopsies.

Bone·2025
Same author

[Tinnitus following treatment with ultrasonic scaler].

Nederlands tijdschrift voor tandheelkunde·2023
Same author

Modeling-based bone formation transforms trabeculae to cortical bone in the sclerotic areas in Buschke-Ollendorff syndrome. A case study of two females with LEMD3 variants.

Bone·2020
Same author

A phase II clinical trial does not show that high dose simvastatin has beneficial effect on markers of bone turnover in multiple myeloma.

Hematological oncology·2008
Same author

Pulse treatment with the proteasome inhibitor bortezomib inhibits osteoclast resorptive activity in clinically relevant conditions.

Leukemia research·2008
Same author

Biochemical markers of type II collagen breakdown and synthesis are positioned at specific sites in human osteoarthritic knee cartilage.

Osteoarthritis and cartilage·2007

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Neural Networks

Background:

  • Multilayer perceptrons (MLPs) are widely used in machine learning.
  • Existing MLP construction techniques face challenges like premature network growth termination.
  • There is a need for improved MLP training methods that enhance generalization performance.

Purpose of the Study:

  • To introduce Dynamic Multilayer Perceptron 3 (DMP3), a novel constructive training algorithm for MLPs.
  • To demonstrate the advantages of DMP3 over existing MLP construction techniques.
  • To evaluate the generalization performance of DMP3 on real-world datasets.

Main Methods:

  • DMP3 constructs MLPs by incrementally adding network elements of varying complexity.
  • The growth of the network is guided by information gain, not error minimization.

Related Experiment Videos

  • DMP3's performance is compared against several established machine learning and neural network algorithms.

    • Main Results:

    • DMP3 demonstrated superior average generalization performance compared to other algorithms tested.
    • The use of information gain effectively guided network growth and avoided premature dead ends.
    • The study analyzed the key factors contributing to DMP3's enhanced performance.

    Conclusions:

    • DMP3 offers a more effective approach to constructing MLPs with improved generalization capabilities.
    • Information gain is a valuable criterion for guiding constructive neural network training.
    • DMP3 represents a significant advancement in the field of neural network learning algorithms.