Finite Mixture Models

Course Code : IA343

Time Hours : 18 hours

Time Periods : not available yet

Lecturer : Mohamed Nadif, Professor, mohamed.nadif [at] parisdescartes.fr

Objective

The goal is to present the finite mixture models particularly in unsupervised learning. Applications in different fields such as bioinformatics, text mining, web mining, image and speech processingwill help to demonstrate the usefulness of this approach that has become classic.

Acquired Skills

In which contexts we can use the mixture model approach and the different variants of the EM algorithm.

Contents

  • Overview of classical approaches
  • Finite mixture models in different fields
  • Maximum likelihood approach : the EM algorithm and applications
  • Classification maximum likelihood approach : the EM algorithm and applications
  • Gaussian, Bernoulli, multinomial and Von Mises-Fisher models
  • Interpretation of classical criteria
  • Different variants of EM
  • Selection of the model
  • Hidden Markov Models

References

  • Finite mixture models, Geoffrey McLachlan and David Peel, Wiley (2000)
  • Data Analysis, Gérard Govaert, ISTE Ltd and John Wiley & Sons Inc (5 août 2009)
  • The EM Algorithm and Extensions, Geoffrey McLachlan and Thriyambakam Krishnan, Wiley (1996)