Projection Pursuit Fitting Gaussian Mixture Models
Gaussian mixture models (GMMs) are widely used to model complex distributions. Usually the parameters of the GMMs are determined in a maximum likelihood (ML) framework. A practical deficiency of ML fitting of the GMMs is the poor performance when dealing with high-dimensional data since a large sample size is needed to match the numerical accuracy that is possible in low dimensions. In this paper we propose a method for fitting the GMMs based on the projection pursuit (PP) strategy. By means of simulations we show that the proposed method outperforms ML fitting of the GMMs for small sizes of training sets.
KeywordsBayesian Information Criterion Gaussian Mixture Model Latent Variable Model Training Sample Size Gaussian Mixture Model Model
- 4.Bishop, C. M.: Neural Networks for Pattern Recognition. Oxford University Press Inc., New York (1995)Google Scholar
- 5.Bishop, C. M.: Latent variable models. In: Jordan, M. I. (ed.): Learning in Graphical Models. The MIT Press, London (1999) 371–403Google Scholar
- 10.Moerland, P.: A comparison of mixture models for density estimation. In: Proceedings of the International Conference on Artificial Neural Networks (1999)Google Scholar
- 13.Wand, M. P., Jones, M. C.: Kernel Smoothing. Charman & Hall/CRC (1995)Google Scholar