Non-linear feature extraction by the coordination of mixture models
Abstract
We present a method for non-linear data projection that offers non-linear versions of Principal Component Analysis and Canonical Correlation Analysis. The data is accessed through a probabilistic mixture model only, therefore any mixture model for any type of data can be plugged in. Gaussian mixtures are one example, but mixtures of Bernoulli's to model discrete data might be used as well. The algorithm minimizes an objective function that exhibits one global optimum that can be found by finding the eigenvectors of some matrix. Experimental results on toy data and real data are provided.
Fichier principal
verbeek03asci.pdf (1.3 Mo)
Télécharger le fichier
VVK03b.png (79.71 Ko)
Télécharger le fichier
Origin : Files produced by the author(s)
Format : Figure, Image
Loading...