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

A cost-function approach to rival penalized competitive learning (RPCL).

Jinwen Ma1, Taijun Wang

  • 1Department of Information Science, School of Mathematical Sciences, Peking University, Beijing, China. jwma@math.pku.edu.cn

IEEE Transactions on Systems, Man, and Cybernetics. Part B, Cybernetics : a Publication of the IEEE Systems, Man, and Cybernetics Society
|August 15, 2006
PubMed
Summary
This summary is machine-generated.

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

Construction of a novel S-scheme AgVO<sub>3</sub>/CaIn<sub>2</sub>S<sub>4</sub> heterostructure for efficient photodegradation of tetracycline hydrochloride.

Dalton transactions (Cambridge, England : 2003)·2024
Same author

A 3D hierarchical TiO<sub>2</sub>/CaIn<sub>2</sub>S<sub>4</sub>/C<sub>3</sub>N<sub>4</sub>arrays photoanode with dual-heterojunction for enhanced photoelectrochemical performance.

Nanotechnology·2024
Same author

Self-Attention-Based Deep Convolution LSTM Framework for Sensor-Based Badminton Activity Recognition.

Sensors (Basel, Switzerland)·2023
Same author

Distributed search and fusion for wine label image retrieval.

PeerJ. Computer science·2022
Same author

Fabrication of 3D hierarchical Fe<sub>2</sub>O<sub>3</sub>/SnO<sub>2</sub>photoanode for enhanced photoelectrochemical performance.

Nanotechnology·2022
Same author

Prevalence and relevant factors of positive RF in brucellosis patients with arthralgia.

PLoS neglected tropical diseases·2021

Distance-sensitive Rival Penalized Competitive Learning (DSRPCL) offers a mathematically grounded approach to clustering unlabeled data. This method ensures correct cluster numbers are automatically selected and weight vectors converge to cluster centers.

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Data Science

Background:

  • Rival Penalized Competitive Learning (RPCL) is a heuristic clustering tool for data with an unknown number of clusters.
  • RPCL lacks a robust mathematical theory for its convergence behavior, limiting its theoretical understanding and application.
  • Addressing this gap is crucial for advancing unsupervised learning techniques.

Purpose of the Study:

  • To develop a mathematical theory for RPCL convergence using a cost-function approach.
  • To introduce and analyze Distance-Sensitive RPCL (DSRPCL) as a generalized form of RPCL.
  • To demonstrate the effectiveness of DSRPCL in unsupervised learning tasks.

Main Methods:

  • Theoretical analysis of DSRPCL through a cost-function minimization framework.

Related Experiment Videos

  • Investigating the association between DSRPCL and the minimization of a cost function on network weight vectors.
  • Conducting simulation experiments to validate theoretical findings.

    • Main Results:

    • DSRPCL is proven to be associated with minimizing a specific cost function.
    • Local minima in the cost function lead to some weight vectors converging to data hyperspheres, while others diverge.
    • Global minimum convergence in DSRPCL automatically selects the correct number of weight vectors, positioning them at cluster centers.

    Conclusions:

    • DSRPCL provides a theoretically sound method for clustering with automatic cluster number selection.
    • The cost-function approach elucidates the convergence properties of RPCL algorithms.
    • DSRPCL demonstrates practical utility in unsupervised color image segmentation and wine data classification.