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

Approximate maximum entropy joint feature inference consistent with arbitrary lower-order probability constraints:

Miller1, Yan

  • 1Department of Electrical Engineering, The Pennsylvania State University, University Park, PA 16802, USA.

Neural Computation
|September 8, 2000
PubMed
Summary

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

Rovibronic Spectroscopy of MgCH3 Ã2E <-- &Xtilde;2A1 Transition.

Journal of molecular spectroscopy·1999
Same author

Glycosylation of CHO-derived recombinant tPA produced under elevated pCO2

Biotechnology progress·1999
Same author

Rapid reduction of Hg(II) by mercuric ion reductase does not require the conserved C-terminal cysteine pair using HgBr2 as the substrate

Biochemistry·1999
Same author

Spontaneous Emulsification of Oils Containing Hydrocarbon, Nonionic Surfactant, and Oleyl Alcohol.

Journal of colloid and interface science·1999
Same author

Pleistocene extinction of genyornis newtoni: human impact on australian megafauna

Science (New York, N.Y.)·1999
Same author

Geographical variation in male courtship behaviour and sexual isolation in wolf spiders of the genus Schizocosa.

Animal behaviour·1998

We introduce an approximate maximum entropy (ME) learning method for discrete statistical classifiers. This approach offers improved classification performance over existing models while simplifying learning complexity.

Area of Science:

  • Machine Learning
  • Statistical Classification
  • Computational Statistics

Background:

  • Estimating joint probability mass functions (p.m.f.) is key for discrete statistical classifiers.
  • Maximum entropy (ME) models capture complex feature dependencies but face high learning complexity.
  • Bayesian networks (BNs) require explicit conditional independencies, often difficult with limited data.

Purpose of the Study:

  • To develop a tractable approximate maximum entropy (ME) learning method for discrete space statistical classifiers.
  • To improve classification performance compared to existing methods.
  • To enable more general inference and handle missing features.

Main Methods:

  • Proposed an approximate ME method restricting joint p.m.f. support to a subset of the feature space during learning.

Related Experiment Videos

  • Compared classification gains against dependence trees, tree-augmented naive Bayes, BNs trained by Kutato, and multilayer perceptrons.
  • Introduced a novel exact inference method for scenarios with multiple missing features.

    • Main Results:

    • The approximate ME method achieves classification gains over several established models.
    • The method retains tractable learning while incorporating general constraints.
    • Demonstrated potential for extensions to broader inference problems.

    Conclusions:

    • The proposed approximate ME learning method offers a practical solution for discrete statistical classification.
    • This approach balances the power of ME models with computational feasibility.
    • The method shows promise for enhanced classification and inference, including handling missing data.