GMMBAYES - Gaussian Mixture Model Methods



Goal of this project is to study existing and develop new methods for learning Gaussian mixture models. Furthermore, class conditional probability densities constructed by Gaussian mixture models and their usage in classification are considered during the project.

The main practical goal of this project is to implement efficient classification functionality (training and classification) based on statistical theory (e.g., Bayesian inference) and Gaussian mixture model probability densities. The ML parameter estimation will be extensively studied.


Released a patch for GMMBayes ToolBox v1.0 to replace the gmmb_covfixer function with a better one. More on the download page. 9.12.2005

GMMBayes ToolBox v1.0 released 14.4.2005. Final version.

Research Report 95 published, see Other scientific publications, March 2005. Fractile document is removed.

GMMBayes ToolBox v0.3 and preliminary document about fractiles released 3.11.2004.

GMMBayes ToolBox v0.2 released 31.8.2004.

Project documentation released 24.6.2004.

GMMBayes ToolBox v0.1 released.

Project in documentation and testing phase.

Contact information

Pekka Paalanen E-Mail WWW Post-graduate researcher
Joni Kämäräinen E-Mail WWW Project leader, lecturer
Heikki Kälviäinen E-Mail WWW Professor


Source code

  1. GMMBayes Toolbox for Matlab - Gaussian mixture model learning and Bayesian classification.

Links to other resources


Do not hesitate to to contact authors (e-mail, etc.) in order to retrieve copies or reprints of the following publications.

Other publications

  1. Information Technology Project report: report04.pdf (296kiB).
    Erratum: The document contains an incorrect formula for computing complex Gaussian PDF values. The correct formula is:
    1/(piD|S|) e -(x-u)*S-1(x-u)
  2. Corretly projected 3D-plots from the report as 150 dpi bitmap images: correct-3d-graphs.pdf (1.7MiB).