An Effective EM Algorithm for PCA Mixture Model
This paper studied PCA mixture model in high dimensional space. A novel EM learning approach by using perturbation was proposed for the PCA mixture model. Experiments showed the novel perturbation EM algorithm is more effective in learning PCA mixture model than an existing constrained EM algorithm.
KeywordsMixture Model Gaussian Mixture Model Principal Direction High Dimensional Space Neural Computation
- 1.Titterington, D.M., Smith, A.F.M., Makov, U.E.: Statistical analysis of finite mixture distribution. Wiley, New York (1985)Google Scholar
- 7.Verbeek, J.J., Vlassis, N., Kröse, B.: Coordinating mixtures of probabilistic principal component analyzers. Technical report, Computer Science Institute, University of Amsterdam, The Netherlands (Febraury 2002) IAS-UVA-02-01Google Scholar