

I am going to outline directions of my current research very briefly, then to present a new Bayes classifier to separate multidimensional numeric data and describe its properties.
To classify data into two classes by the Fisher linear discriminant analysis it is actually supposed those data to be the result of observation of two points sampled with Gaussian random noise. Consequently, the Fisher classifier is a hyperplane separating the classes.
We suppose our data is a result of sampling two disjoint sets that are observed with some random noise, optionally Gaussian. We offer a Bayes classifier, which is the MAP estimate of the joint density of the data, and describe properties of the classifier on some prior conditions for forms of sets and distributions of the random noise.
As a matter of fact, the classifier is a separating surface.
Page maintenue par
le webmaster
 3 février 2010
© Télécom ParisTech/TSI 19982010
