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

Predictive minimum description length criterion for time series modeling with neural networks

M Lehtokangas1, J Saarinen, P Huuhtanen

  • 1Microelectronics Laboratory, Tampere University of Technology, Finland.

Neural Computation
|April 1, 1996
PubMed
Summary

The predictive minimum description length (PMDL) principle efficiently selects neural network models for nonlinear time series. This method is faster than cross-validation while achieving comparable 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

Rivaroxaban versus aspirin for secondary prevention of ischaemic stroke in patients with cancer: a subgroup analysis of the NAVIGATE ESUS randomized trial.

European journal of neurology·2020
Same author

Randomized clinical trial comparing surgery, endovenous laser ablation and ultrasound-guided foam sclerotherapy for the treatment of great saphenous varicose veins.

The British journal of surgery·2016
Same author

Criteria for minimal model of driven polymer translocation.

Physical review. E, Statistical, nonlinear, and soft matter physics·2014
Same author

Dynamics of polymer ejection from capsid.

Physical review. E, Statistical, nonlinear, and soft matter physics·2014
Same author

Sedimentation of knotted polymers.

Physical review. E, Statistical, nonlinear, and soft matter physics·2013
Same author

Rapid action in the Palaeogene, the relationship between phenotypic and taxonomic diversification in Coenozoic mammals.

Proceedings. Biological sciences·2012

Area of Science:

  • Computational neuroscience
  • Machine learning
  • Time series analysis

Background:

  • Nonlinear time series modeling often utilizes multilayer perceptron networks.
  • Model selection, determining network size and complexity, is a critical challenge.

Purpose of the Study:

  • To apply the predictive minimum description length (PMDL) principle for neural network model selection in nonlinear time series.
  • To evaluate the effectiveness of PMDL against traditional cross-validation.

Main Methods:

  • Utilized the PMDL principle as a minimization criterion to determine optimal input and hidden units in neural networks.
  • Conducted three time series modeling experiments to assess PMDL.
  • Compared PMDL performance with the cross-validation technique.

Related Experiment Videos

Main Results:

  • PMDL and cross-validation yielded similar results regarding model complexity.
  • The PMDL method demonstrated a two-fold increase in computational speed compared to cross-validation.
  • This speed improvement is significant for time-consuming model selection processes.

Conclusions:

  • The PMDL principle offers an efficient and effective approach for model selection in nonlinear time series modeling.
  • PMDL provides a valuable alternative to cross-validation, particularly when computational time is a concern.