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

Feature decomposition architectures for neural networks: algorithms, error bounds, and applications.

Haiying Wang1, Snehasis Mukhopadhyay, Shiaofen Fang

  • 1Department of Computer Information Science, Indiana University, Indianapolis, IN 46202, USA.

International Journal of Neural Systems
|February 20, 2002
PubMed
Summary

Modular neural networks offer advantages over monolithic designs. This study introduces feature decomposition models and algorithms, demonstrating competitive performance in pattern recognition and modeling 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 author

Findings of a videofluoroscopic swallowing study in patients with dysphagia.

Frontiers in neurology·2023
Same author

Analysis of changes in the spatiotemporal characteristics of impervious surfaces and their influencing factors in the Central Plains Urban Agglomeration of China from 2000 to 2018.

Heliyon·2023
Same author

Nanoparticle-mediated synergistic anticancer effect of ferroptosis and photodynamic therapy: Novel insights and perspectives.

Asian journal of pharmaceutical sciences·2023
Same author

Review on Processing Methods of Toxic Chinese Materia Medica and the Related Mechanisms of Action.

The American journal of Chinese medicine·2023
Same author

Pro‑angiogenic activity of salvianolate and its potential therapeutic effect against acute cerebral ischemia.

Experimental and therapeutic medicine·2023
Same author

CXCL12-CXCR4/CXCR7 Axis in Cancer: from Mechanisms to Clinical Applications.

International journal of biological sciences·2023

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Computer Science

Background:

  • Modular neural networks are gaining traction due to their advantages over single, large monolithic networks.
  • These systems offer benefits in scalability, interpretability, and potentially improved performance.

Purpose of the Study:

  • To propose novel feature decomposition models for modular neural networks.
  • To introduce an algorithm for partitioning input features based on training data.
  • To evaluate the performance of these networks against monolithic architectures.

Main Methods:

  • Development of two feature decomposition models: parallel and tandem.
  • Introduction of a feature decomposition algorithm to partition input space.
  • Empirical comparison of feature decomposition networks with monolithic networks on benchmark problems.

Related Experiment Videos

Main Results:

  • The proposed feature decomposition algorithm effectively partitions input features using only training data.
  • Theoretical analysis shows the approximation error due to decomposition can be bounded.
  • Experimental results demonstrate competitive performance of feature decomposition networks compared to monolithic networks.

Conclusions:

  • Feature decomposition networks provide a viable and effective alternative to monolithic neural networks.
  • The proposed methods offer a data-driven approach to modular network design.
  • These findings support the continued exploration of modular architectures in machine learning.