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

Nonlinear gated experts for time series: discovering regimes and avoiding overfitting

A S Weigend1, M Mangeas, A N Srivastava

  • 1Department of Computer Science, University of Colorado, Boulder, 80309-0430, USA.

International Journal of Neural Systems
|December 1, 1995
PubMed
Summary

Gated experts effectively address nonstationarity and overfitting in time series analysis by using a gating network to partition input data, leading to improved prediction accuracy and reduced model complexity.

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

Impact of lockdown on cardiovascular disease hospitalizations in a Zero-COVID-19 country.

Public health·2023
Same author

Exploring nature's bounty: identification of <i>Withania somnifera</i> as a promising source of therapeutic agents against COVID-19 by virtual screening and <i>in silico</i> evaluation.

Journal of biomolecular structure & dynamics·2020
Same author

Anticancer Potential of Dietary Natural Products: A Comprehensive Review.

Anti-cancer agents in medicinal chemistry·2019
Same author

Synergic effects of cancer stem cells markers, CD44 and embryonic stem cell transcription factor Nanog, on bladder cancer prognosis.

British journal of biomedical science·2019
Same author

Correlation of Hormone Receptor and Human Epidermal Growth Factor Receptor-2/neu Expression in Breast Cancer with Various Clinicopathologic Factors.

Indian journal of medical and paediatric oncology : official journal of Indian Society of Medical & Paediatric Oncology·2018
Same author

Pathology of Ringworm and Allied Skin Diseases.

The Indian medical gazette·2017

Area of Science:

  • Machine Learning
  • Time Series Analysis
  • Statistical Modeling

Background:

  • Real-world systems analysis faces challenges with nonstationarity and overfitting, especially in noisy data.
  • Traditional methods struggle to adapt to dynamic system changes and complex noise patterns.

Purpose of the Study:

  • To introduce and evaluate a novel approach using gated experts for improved analysis and prediction of complex time series.
  • To address limitations of existing models in handling nonstationarity and overfitting.

Main Methods:

  • Developed a system of gated experts, comprising a nonlinear gating network and multiple nonlinear competing experts.
  • Experts predict conditional means and adapt their prediction width to local noise levels.
  • Gating network learns to assign probabilities to experts based on input data, enabling soft partitioning of the input space.

Related Experiment Videos

Main Results:

  • The gating network successfully identified distinct regimes within the time series data.
  • Expert-associated widths proved crucial for segmentation and characterizing subprocesses.
  • Gated experts demonstrated reduced overfitting compared to single networks by matching local model complexity to data complexity.

Conclusions:

  • Gated experts offer a robust solution for modeling nonstationary and noisy time series data.
  • The adaptive nature of experts and the input-driven gating mechanism enhance predictive performance.
  • This approach provides a more nuanced understanding of system dynamics by segmenting data into distinct regimes.